@paigy/mcp 0.40.42 → 0.40.44

package/dist/{chunk-NNB3HQTF.js → chunk-LRROOCLY.js}

@@ -1,23 +1,24 @@
 // ../../packages/schema/dist/index.js
-import { z as z13 } from "zod";
+import { z as z14 } from "zod";
 import { z } from "zod";
-import { z as z3 } from "zod";
+import { z as z4 } from "zod";
 import { z as z2 } from "zod";
-import { z as z6 } from "zod";
+import { z as z3 } from "zod";
+import { z as z7 } from "zod";
 import { zodToJsonSchema } from "zod-to-json-schema";
-import { z as z4 } from "zod";
 import { z as z5 } from "zod";
-import { z as z7 } from "zod";
-import { zodToJsonSchema as zodToJsonSchema2 } from "zod-to-json-schema";
+import { z as z6 } from "zod";
 import { z as z8 } from "zod";
-import { zodToJsonSchema as zodToJsonSchema3 } from "zod-to-json-schema";
+import { zodToJsonSchema as zodToJsonSchema2 } from "zod-to-json-schema";
 import { z as z9 } from "zod";
-import { zodToJsonSchema as zodToJsonSchema4 } from "zod-to-json-schema";
+import { zodToJsonSchema as zodToJsonSchema3 } from "zod-to-json-schema";
 import { z as z10 } from "zod";
-import { zodToJsonSchema as zodToJsonSchema5 } from "zod-to-json-schema";
+import { zodToJsonSchema as zodToJsonSchema4 } from "zod-to-json-schema";
 import { z as z11 } from "zod";
-import { zodToJsonSchema as zodToJsonSchema6 } from "zod-to-json-schema";
+import { zodToJsonSchema as zodToJsonSchema5 } from "zod-to-json-schema";
 import { z as z12 } from "zod";
+import { zodToJsonSchema as zodToJsonSchema6 } from "zod-to-json-schema";
+import { z as z13 } from "zod";
 import { zodToJsonSchema as zodToJsonSchema7 } from "zod-to-json-schema";
 var OPTIONS_MIN = 2;
 var OPTIONS_MAX = 6;
@@ -97,29 +98,43 @@ function capRefusals(text, units2, field) {
 function refuseOverCaps(text, units2, field, ctx) {
   for (const { path, message, ...params } of capRefusals(text, units2, field)) ctx.addIssue({ code: z2.ZodIssueCode.custom, path, message, params });
 }
+var SHIPPED = ["phone", "calls", "server", "store", "agents"];
+var CAUSES_MAX = 10;
+var shipped = z3.enum(SHIPPED, {
+  errorMap: () => ({ message: `shipped_unknown: one of ${SHIPPED.join(", ")}.` })
+}).optional().describe(
+  "Only when this reports a result that LANDED somewhere the person can see it: phone (an app update reached their phone), calls (the voice bot), server, store (the App Store) or agents (the MCP or SDK). Leave it out for work in progress."
+);
+var causedBy = z3.array(
+  z3.string().trim().regex(/^[0-9a-fA-F-]{8,36}$/, "cause_not_found: an Entry id, or the short id get_goals prints for the person's words.")
+).min(1).max(CAUSES_MAX, `too_many_causes: at most ${CAUSES_MAX}.`).optional().describe(
+  `The person's own words this acts on: the ids (or the short ids get_goals prints) of their answers or notes that led to it, at most ${CAUSES_MAX}. Their app shows each result under the words that caused it, so name every one this follows from. An id that is not their words is refused by name (cause_not_found, cause_ambiguous) and nothing is written.`
+);
 var COMPLETION_MESSAGE_MAX = 50;
-var CompletionSchema = z3.object({
-  goalId: z3.string().uuid().describe("The existing Goal to complete."),
-  message: z3.string().trim().min(1).max(COMPLETION_MESSAGE_MAX, "completion_message_too_long: at most 50 characters; put detail in units.").describe("Completion headline, at most 50 characters."),
-  units: z3.array(UnitInputSchema.extend({
-    title: z3.string().trim().min(1).max(UNIT_TITLE_MAX),
-    body: z3.string().trim().min(1).max(UNIT_BODY_MAX)
+var CompletionSchema = z4.object({
+  goalId: z4.string().uuid().describe("The existing Goal to complete."),
+  message: z4.string().trim().min(1).max(COMPLETION_MESSAGE_MAX, "completion_message_too_long: at most 50 characters; put detail in units.").describe("Completion headline, at most 50 characters: the one thing that would go wrong or stay unclear later if they missed it."),
+  units: z4.array(UnitInputSchema.extend({
+    title: z4.string().trim().min(1).max(UNIT_TITLE_MAX),
+    body: z4.string().trim().min(1).max(UNIT_BODY_MAX)
   })).max(UNITS_MAX).optional().describe("Suggest initial state, changes, current state with verification and limits, and next steps / leftover work with its owner when useful. One idea per unit; omit unnecessary sections."),
-  reply: z3.string().trim().min(1).max(OPTION_LABEL_MAX).describe("One acknowledgement in the person's voice, at most 50 characters. Acknowledgement alone requests no follow-up and does not claim they tested the work."),
-  options: z3.array(z3.object({
-    label: z3.string().trim().min(1).max(OPTION_LABEL_MAX),
-    hint: z3.string().trim().min(1).max(OPTION_HINT_MAX)
-  }).strict()).max(5).optional().describe("Optional next steps the person may initiate: label (50 characters) and consequence in hint (200). Selecting one acknowledges the report and requests that step from its author; it never reopens this Goal.")
+  reply: z4.string().trim().min(1).max(OPTION_LABEL_MAX).describe("One acknowledgement in the person's voice, at most 50 characters. Acknowledgement alone requests no follow-up and does not claim they tested the work."),
+  options: z4.array(z4.object({
+    label: z4.string().trim().min(1).max(OPTION_LABEL_MAX),
+    hint: z4.string().trim().min(1).max(OPTION_HINT_MAX)
+  }).strict()).max(5).optional().describe("Optional next steps the person may initiate: label (50 characters) and consequence in hint (200). Selecting one acknowledges the report and requests that step from its author; it never reopens this Goal."),
+  causedBy,
+  shipped
 }).strict();
-var CompleteChangeSchema = CompletionSchema.extend({ kind: z3.literal("complete") });
+var CompleteChangeSchema = CompletionSchema.omit({ causedBy: true, shipped: true }).extend({ kind: z4.literal("complete") });
 var ASKING_RULES = `HOW TO ASK:
 1. One question per object in \`questions\`, so each is answered on its own.
 2. \`goalId\` names the existing Goal it is about (create one with manage_goals first); the answer comes back there.
-3. \`workItBlocks\` only when you are truly blocked: it asks for a call.
+3. They asked you to call? \`userExplicitlyRequested:"call"\`; \`workItBlocks\` rings when blocked.
 4. No while-loops. While any Call is open, even one your question joined from their items, keep receiving through empty windows; otherwise stop at one: the answer reaches you on the Goal. Yield only with a listener or wakeup.
 5. One question, one row: never re-send one still waiting; take back one that no longer matters with \`withdrawQuestionIds\`.
 6. Read the Goal before asking; never re-ask what was answered or shipped. \`replied\` means they chose no option: it is still yours to settle.
-7. Give options when you have them. Each label names its choice in full (never "Option 2"); never add "Other". \`pickMode\`: one (default), many, rank.
+7. Give options when you have them. Each label names its choice in full (never "Option 2"); never add "Other".
 8. A diagnosis says when, why and how it happens, then proposes one fix.
 9. Give a report its \`reply\`: one line in THEIR voice for THAT report, so you learn whether they verified it or only received it. Required when it is delivered or its Goal is done.`;
 function draft2020(node) {
@@ -151,25 +166,68 @@ function mcpInputSchema(s) {
   delete schema.$schema;
   return draft2020(schema);
 }
-var questionHandle = z4.string().trim().regex(/^[0-9a-fA-F-]{8,36}$/, "a question id, whole or its first eight characters");
-var NodeRef = z4.object({
-  type: z4.enum(["goal", "question"]),
-  id: z4.string().uuid()
+var questionHandle = z5.string().trim().regex(/^[0-9a-fA-F-]{8,36}$/, "a question id, whole or its first eight characters");
+var NodeRef = z5.object({
+  type: z5.enum(["goal", "question"]),
+  id: z5.string().uuid()
 }).strict().describe("A Goal or a Question, by its id.");
-var BlockedAction = z4.enum(["start", "complete", "answer"]).describe("What waits: starting or completing a Goal, or answering a Question.");
-var reason = z4.string().trim().min(1).max(500).describe("Why, as an explanation, never as policy.");
+var BlockedAction = z5.enum(["start", "complete", "answer"]).describe("What waits: starting or completing a Goal, or answering a Question.");
+var reason = z5.string().trim().min(1).max(500).describe("Why, as an explanation, never as policy.");
 var actionFits = (b) => b.blocked.type === "goal" === (b.action !== "answer");
 var ACTION_FITS = "a goal is blocked from start or complete; a question from answer";
-var userExplicitlyRequested = z4.enum(["any", "call"]).optional().describe(
-  `Only what the person said about THIS message, in so many words: "any" when they asked you to send it to them ("send it to my phone", "message me the result"), "call" when they asked to be called about it. It is then delivered (a call rings for the whole contact, and carries everything waiting on them); omitted, an update is only recorded as the Goal's progress unless the person needs it. Never set it to make a message louder.`
+var userExplicitlyRequested = z5.enum(["any", "call"]).optional().describe(
+  'What the person asked for about THIS message, in so many words: "any" when they asked you to send it to them ("send it to my phone", "message me the result"), "call" when they asked to be called about it. It is then delivered (a call rings for the whole contact, and carries everything waiting on them); omitted, an update is only recorded as the Goal\'s progress unless the person needs it. ALSO "any" whenever this message directly answers a question they asked you, wherever they asked it -- on a call, in a reply, in passing: their question is the request, so it is delivered every time, even with no pending question of theirs on file and even if they never said to send it. A correction of something you or another agent told them is such an answer. Never set it to make a message louder. This is the field that rings a phone: "call" here is how you call them when they asked you to, and `help` has the whole mechanism.'
 );
-var units = z4.array(UnitInputSchema).optional().describe(
-  `The context this message needs, as up to ${UNITS_MAX} units in reading order, each {title, body}: a title of at most ${UNIT_TITLE_MAX} characters and a body of at most ${UNIT_BODY_MAX}. One idea per unit. Leave it out when the main text says it all. Over a cap is refused by name (ask_too_long, too_many_units, unit_title_too_long, unit_body_too_long) and nothing is sent.`
+var units = z5.array(UnitInputSchema).optional().describe(
+  `The context this message needs, as up to ${UNITS_MAX} units in reading order, each {title, body}: a title of at most ${UNIT_TITLE_MAX} characters and a body of at most ${UNIT_BODY_MAX}. One idea per unit: context they may open, never something they must hear. Leave it out when the main text says it all. Over a cap is refused by name (ask_too_long, too_many_units, unit_title_too_long, unit_body_too_long) and nothing is sent.`
 );
-var UpdateInputSchema = z4.object({
-  message: z4.string().trim().min(1).describe(`The one point, at most ${ASK_MAX} characters; the context goes in units.`),
-  goalIds: z4.array(z4.string().uuid()).min(1).max(10).describe("The Goals it is about, each one that exists: the one it is mainly about first."),
+var KIND_IDS = { "handing": "none", "replaced-by": "goal", "blocked-by": "node", "asking": "none", "found": "none" };
+var KIND_GRAMMAR = "handing, replaced-by <goalId>, blocked-by <goalId|questionId>, asking, found";
+var uuid = /^[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}$/;
+var handle = /^[0-9a-fA-F-]{8,36}$/;
+function refuseKind(u, ctx) {
+  if (!u.kind) return;
+  const refuse = (refusal, kind, message) => ctx.addIssue({ code: z5.ZodIssueCode.custom, path: ["kind"], params: { refusal, kind }, message });
+  const [name, ...rest] = u.kind.trim().split(/\s+/);
+  const shape = KIND_IDS[name];
+  if (!shape) {
+    refuse("kind_unknown", name, `kind_unknown: "${name}" is not a kind. One of: ${KIND_GRAMMAR}.`);
+    return;
+  }
+  const id2 = rest.join(" ");
+  if (shape === "none" && id2) {
+    refuse("kind_takes_no_id", name, `kind_takes_no_id: "${name}" is the whole kind, so drop the "${id2}" after it.`);
+  }
+  if (shape !== "none" && !(shape === "goal" ? uuid.test(id2) : handle.test(id2))) {
+    refuse("kind_needs_id", name, shape === "goal" ? `kind_needs_id: "${name}" names the Goal it is about, as "${name} <goalId>".` : `kind_needs_id: "${name}" names the Goal or Question it is about, as "${name} <goalId|questionId>".`);
+  }
+  if (name === "asking") {
+    refuse("kind_is_a_question", name, 'kind_is_a_question: "asking" owes them an answer, so send it in `questions`, where it opens one.');
+  }
+  if (u.userExplicitlyRequested) {
+    refuse("kind_not_for_the_person", name, `kind_not_for_the_person: "${name}" is written for the other agents on the Goal, and userExplicitlyRequested delivers these words to the person. Split it: one message for them, one for the agent.`);
+  }
+}
+var UpdateInputSchema = z5.object({
+  // ONE LOAD-BEARING POINT PER REPORT (owner, 2026-10-27 14:10 on call, Goal 1b1750e9, settling
+  // #3633's pacing: "one point at a time is better, but make sure that it's [load-]bearing ... if
+  // there is a lot of key points that I need to be aware of ... meaning something bad would happen
+  // if I didn't know them or I wouldn't be able to understand something in the future ... They
+  // should go separately. Basically, it should be like one load-bearing thing per report").
+  //
+  // The call's half of that ruling shipped in #3771: the talker settles an item only once the
+  // transcript shows he heard its load-bearing point. This is the author's half, and without it the
+  // talker has to guess which of a report's parts was load-bearing. Measured 2026-11-01 on prod:
+  // 458 of 866 reports in 21 days carried two or more units, and since the pacing shipped on 10-23
+  // only 18 of 72 units that reached a call were ever said -- so a point written as a unit has about
+  // a one-in-four chance of being heard, and one ok after the headline settles the whole report.
+  // Hence the wording, not a `mustHear` flag (the-heard-record.md P3): he asked for less mechanism
+  // and more judgment from the agent, so a point he must not miss is simply its own report.
+  message: z5.string().trim().min(1).describe(`The one point, at most ${ASK_MAX} characters: the one thing that would go wrong or stay unclear later if they missed it, said as what changed for them, in their words ("The screen reader speaks Spanish too", not "Merged #3896"). A second point like that is its own update; the rest goes in units.`),
+  goalIds: z5.array(z5.string().uuid()).min(1).max(10).describe("The Goals it is about, each one that exists: the one it is mainly about first."),
   units,
+  causedBy,
+  shipped,
   // A REPORT KEEPS ITS TAP (#2799, #3260; owner, 2026-10-08, settling #2776: "keep the reports open
   // until the user acknowledges them"). A report asks nothing, so it opens no Question -- but it
   // still waits to be acknowledged, and it carries the ONE tap that closes it. Never a decision: the
@@ -190,23 +248,30 @@ var UpdateInputSchema = z4.object({
   // `options` and this field are one job: the list is deleted, not deprecated (ONE JOB, ONE
   // MECHANISM). It rides the door as the report's single stored option (`storedOption`,
   // `apps/api/src/goal/router.ts`), so no reader below the door learns a second shape.
-  reply: z4.string().trim().min(1).describe(
+  reply: z5.string().trim().min(1).describe(
     `What this report's button says instead of "Got it", at most ${OPTION_LABEL_MAX} characters, written in THEIR voice for THIS report ("Sounds good, I'll test later") -- the care you give a question's options, so what they tap tells you whether they verified your work or only received it. One tap sends exactly these words back to you. Words that would CHANGE what happens next are a question, not a reply to a report: it still asks nothing and still owes you no answer. The reply is required for requested delivery and completed Goals; ordinary progress may omit it.`
   ).optional(),
+  // THE KIND, WHICH WAS THE FIRST LINE OF THE MESSAGE UNTIL NOW (`KIND_IDS` above). Optional on
+  // purpose: every agent writing updates today sends none, and a door that refused them would turn a
+  // convention nobody was obliged to follow into a 400 on live traffic.
+  kind: z5.string().trim().min(1).optional().describe(
+    `What an agent writing on ANOTHER agent's Goal expects of its reader, as one token: ${KIND_GRAMMAR}. \`handing\` the work is yours now, nothing is asked; \`replaced-by <goalId>\` this Goal is dead and that one carries its work; \`blocked-by <goalId|questionId>\` I cannot go on until that moves, no action unless you own it; \`found\` evidence on your Goal, nothing is asked and nothing has changed. \`asking\` is a question, so it belongs in \`questions\`, not here. It is written for the other agents on the Goal, never the person, so it does not go with userExplicitlyRequested; leave it out for your own progress. Refused by name (kind_unknown, kind_needs_id, kind_takes_no_id, kind_is_a_question, kind_not_for_the_person).`
+  ),
   userExplicitlyRequested
 }).strict().superRefine((u, ctx) => {
   if (u.userExplicitlyRequested && !u.reply) {
     ctx.addIssue({
-      code: z4.ZodIssueCode.custom,
+      code: z5.ZodIssueCode.custom,
       path: ["reply"],
       params: { refusal: "reply_required" },
       message: "reply_required: an update requested for delivery needs one reply, written in the person\u2019s voice for this report."
     });
   }
   refuseOverCaps(u.message, u.units, "message", ctx);
+  refuseKind(u, ctx);
   if (u.reply && u.reply.length > OPTION_LABEL_MAX) {
     ctx.addIssue({
-      code: z4.ZodIssueCode.custom,
+      code: z5.ZodIssueCode.custom,
       path: ["reply"],
       params: { refusal: "reply_too_long", length: u.reply.length, cap: OPTION_LABEL_MAX },
       message: `reply_too_long: the reply is ${u.reply.length} characters; the cap is ${OPTION_LABEL_MAX}. It is a button's words, so say it in a breath.`
@@ -214,31 +279,31 @@ var UpdateInputSchema = z4.object({
   }
 });
 var questionFields = {
-  question: z4.string().trim().min(1).describe(
+  question: z5.string().trim().min(1).describe(
     `ONE question, in at most ${ASK_MAX} characters, and only what is needed to answer it; the context goes in units. Several questions are several objects, one each: an answer settles the one question it was given, so a person who answers the part of a bundled question that interested them settles nothing and is asked again. Separate questions cost nothing: ANY contact for a person already on a call JOINS that call, so several arrive as one call.`
   ),
-  goalId: z4.string().uuid().describe("The Goal the question is about, which must exist (create it with manage_goals first). Its answer comes back there."),
+  goalId: z5.string().uuid().describe("The Goal the question is about, which must exist (create it with manage_goals first). Its answer comes back there."),
   units,
-  options: z4.array(OptionInputSchema).min(1).max(6).optional().describe("The choices, when you have them. Each label stands on its own and differs from the others (duplicate_option_label), at most 50 characters; a hint, at most 200. They can always answer in their own words, so no 'Other' or 'None of these'."),
+  options: z5.array(OptionInputSchema).min(1).max(6).optional().describe("The choices, when you have them. Each label stands on its own and differs from the others (duplicate_option_label), at most 50 characters; a hint, at most 200. They can always answer in their own words, so no 'Other' or 'None of these'."),
   // HOW THE OPTIONS ARE ANSWERED, SAID BY THE AGENT (owner, 2026-10-02: "one and many makes sense";
   // brain_prompts.md §1: structured agent questions need no model). The app draws all three.
-  pickMode: z4.enum(["one", "many", "rank"]).optional().describe(
+  pickMode: z5.enum(["one", "many", "rank"]).optional().describe(
     'How the options are answered: "one" (the default) a pick, for alternatives; "many" a checklist, for independent options they may want several of; "rank" an order of all of them. Only with options; "many" and "rank" need at least two (too_few_options).'
   ),
   // WHAT THE QUESTION HOLDS UP (brain_prompts.md §2.1): replaces `waiting: 'hard'`. A question that
   // blocks work asks for a call (the person's settings and ring cooldowns still decide).
-  workItBlocks: z4.array(z4.object({ blocked: NodeRef, action: BlockedAction, reason: reason.optional() }).strict().refine(actionFits, ACTION_FITS)).min(1).max(10).optional().describe(
+  workItBlocks: z5.array(z5.object({ blocked: NodeRef, action: BlockedAction, reason: reason.optional() }).strict().refine(actionFits, ACTION_FITS)).min(1).max(10).optional().describe(
     "The work that cannot go on until this is answered: a Goal's start or completion, or another Question's answer. Only for work truly blocked; it asks for a call. Leave it out for a question you can keep working around."
   ),
   userExplicitlyRequested
 };
 var refineQuestion = (q, ctx) => {
   refuseOverCaps(q.question, q.units, "question", ctx);
-  if (q.pickMode && !q.options) ctx.addIssue({ code: z4.ZodIssueCode.custom, path: ["pickMode"], message: "pickMode comes only with options" });
+  if (q.pickMode && !q.options) ctx.addIssue({ code: z5.ZodIssueCode.custom, path: ["pickMode"], message: "pickMode comes only with options" });
   refuseThinOptions(q, ctx);
   const ids = (q.options ?? []).map((o) => o.id);
   if (ids.some((id2) => id2) && (ids.some((id2) => !id2) || new Set(ids).size !== ids.length)) {
-    ctx.addIssue({ code: z4.ZodIssueCode.custom, path: ["options"], message: "give every option a distinct id, or none" });
+    ctx.addIssue({ code: z5.ZodIssueCode.custom, path: ["options"], message: "give every option a distinct id, or none" });
   }
 };
 var MIN_LIST_OPTIONS = 2;
@@ -246,7 +311,7 @@ function refuseThinOptions(q, ctx) {
   const options = q.options ?? [];
   if ((q.pickMode === "many" || q.pickMode === "rank") && options.length < MIN_LIST_OPTIONS) {
     ctx.addIssue({
-      code: z4.ZodIssueCode.custom,
+      code: z5.ZodIssueCode.custom,
       path: ["options"],
       params: { refusal: "too_few_options", length: options.length, min: MIN_LIST_OPTIONS },
       message: `too_few_options: pickMode "${q.pickMode}" needs at least ${MIN_LIST_OPTIONS} options; this question has ${options.length}. Add the other choices, or leave pickMode out.`
@@ -261,67 +326,67 @@ function refuseThinOptions(q, ctx) {
       return;
     }
     ctx.addIssue({
-      code: z4.ZodIssueCode.custom,
+      code: z5.ZodIssueCode.custom,
       path: ["options", index, "label"],
       params: { refusal: "duplicate_option_label", index, first },
       message: `duplicate_option_label: options[${index}] repeats the label of options[${first}]. Each option names a different choice.`
     });
   });
 }
-var QuestionInputSchema = z4.object({
-  questionId: z4.string().uuid().optional().describe("Optional: your own stable id for this question (a UUID), so a resend is the same question. Minted for you when omitted; the result returns it."),
+var QuestionInputSchema = z5.object({
+  questionId: z5.string().uuid().optional().describe("Optional: your own stable id for this question (a UUID), so a resend is the same question. Minted for you when omitted; the result returns it."),
   ...questionFields
 }).strict().superRefine(refineQuestion);
-var QuestionWireSchema = z4.object({ questionId: z4.string().uuid(), ...questionFields }).strict().superRefine(refineQuestion);
-var AnswerInputSchema = z4.object({
+var QuestionWireSchema = z5.object({ questionId: z5.string().uuid(), ...questionFields }).strict().superRefine(refineQuestion);
+var AnswerInputSchema = z5.object({
   questionId: questionHandle.describe("The Question you answer: its id as the event or the conversation shows it (8 characters or whole)."),
-  answer: z4.object({
-    text: z4.string().trim().min(1).describe(`Your answer in your own words, at most ${ASK_MAX} characters; the context goes in units.`),
-    selectedOptionIds: z4.array(z4.string().trim().min(1)).min(1).max(6).optional().describe("The ids of the options you choose, when the Question offered options (in order, for a ranking).")
+  answer: z5.object({
+    text: z5.string().trim().min(1).describe(`Your answer in your own words, at most ${ASK_MAX} characters; the context goes in units.`),
+    selectedOptionIds: z5.array(z5.string().trim().min(1)).min(1).max(6).optional().describe("The ids of the options you choose, when the Question offered options (in order, for a ranking).")
   }).strict(),
   units
 }).strict().superRefine((a, ctx) => refuseOverCaps(a.answer.text, a.units, "text", ctx));
-var QuestionDependencySchema = z4.object({
+var QuestionDependencySchema = z5.object({
   questionId: questionHandle.describe("A Question you asked."),
-  change: z4.enum(["add", "remove"]),
+  change: z5.enum(["add", "remove"]),
   blocked: NodeRef,
   action: BlockedAction,
   reason: reason.optional()
 }).strict().refine(actionFits, ACTION_FITS);
 var sendFields = {
-  completions: z4.array(CompletionSchema).min(1).max(20).optional().describe(
+  completions: z5.array(CompletionSchema).min(1).max(20).optional().describe(
     "COMPLETIONS close and report in one operation, the same contract as manage_goals kind:complete: closure and the report commit together or neither does. Required children and dependencies must already be finished. options suggest next steps: a tap acknowledges the report and requests that step, and never reopens the Goal (follow-up work needs a new one). No reviewer rejudges it: say only what your evidence supports. Ordinary updates never close a Goal."
   ),
-  updates: z4.array(UpdateInputSchema).min(1).max(20).optional().describe(
-    "News: progress, findings, a result. An update asks nothing. It reaches the person only when they asked for it (userExplicitlyRequested), when it answers something they said, or once its Goal is done; otherwise it is recorded as the Goal's progress, which the person is not told. Every update reaches the other agents that wrote on its Goal in the last hour: to tell another agent something, write it as an update on that agent's Goal."
+  updates: z5.array(UpdateInputSchema).min(1).max(20).optional().describe(
+    "News: progress, findings, a result. An update asks nothing. It reaches the person only when they asked for it (userExplicitlyRequested), when it answers something they said, or once its Goal is done; otherwise it is recorded as the Goal's progress, which the person is not told. Every update reaches the other agents that wrote on its Goal in the last hour: to tell another agent something, write it as an update on that agent's Goal, with its `kind`.\n\nONE LOAD-BEARING POINT PER REPORT: the message is the point they must not miss; a second point that would go wrong or stay unclear later if they missed it is its own update, never a unit."
   ),
-  answers: z4.array(AnswerInputSchema).min(1).max(20).optional().describe("Answers to Questions addressed to you."),
-  withdrawQuestionIds: z4.array(questionHandle).min(1).max(10).optional().describe(
+  answers: z5.array(AnswerInputSchema).min(1).max(20).optional().describe("Answers to Questions addressed to you."),
+  withdrawQuestionIds: z5.array(questionHandle).min(1).max(10).optional().describe(
     "Questions you asked that no longer need an answer: you acted on them yourself, they no longer matter, or you asked wrongly. Each is cancelled, not answered: its card closes and nothing rings for it."
   ),
-  bumpQuestionIds: z4.array(questionHandle).min(1).max(10).optional().describe(
+  bumpQuestionIds: z5.array(questionHandle).min(1).max(10).optional().describe(
     "Your own questions still waiting on the person that you want them to notice again. Each moves back to the top of its card with one new notification; nothing is resent and no second card is made."
   ),
-  questionDependencies: z4.array(QuestionDependencySchema).min(1).max(20).optional().describe(
+  questionDependencies: z5.array(QuestionDependencySchema).min(1).max(20).optional().describe(
     "Add or remove what a Question you asked holds up, after asking it."
   )
 };
-var ContactSchema = z4.object({
+var ContactSchema = z5.object({
   ...sendFields,
-  questions: z4.array(QuestionInputSchema).min(1).max(20).optional().describe("Questions for the person, one per object, each sent exactly as written, with no reading in between."),
+  questions: z5.array(QuestionInputSchema).min(1).max(20).optional().describe("Questions for the person, one per object, each sent exactly as written, with no reading in between."),
   /** RECEIVING (2I, I2; brain_prompts.md §2.2): the events this agent handled. */
-  ackEventIds: z4.array(z4.string().uuid()).min(1).max(100).optional().describe(
+  ackEventIds: z5.array(z5.string().uuid()).min(1).max(100).optional().describe(
     "The exact eventIds of the events you have handled: an acknowledgment (ACK). Reading alone acknowledges nothing, so an event comes back until you acknowledge it; after that the next batch can come. Each recipient acknowledges separately, for itself only. A Question you owe is answered (answers), never acknowledged away."
   ),
-  wait: z4.boolean().optional().describe(
+  wait: z5.boolean().optional().describe(
     "Only controls when this returns. Sending: false (the default) returns at once; true waits up to about 45 seconds for a response to what you sent. Sending nothing: true (the default) waits up to about 45 seconds for anything addressed to you; false returns what is waiting now. It does not mean work is blocked, set urgency, or ask for a call. `waitOutcome` says which happened: `available`, `expired` (nothing arrived; wait again without sending again), or `not_waited`."
   )
 }).strict();
 var sendsAnything = (c) => !!(c.completions?.length || c.updates?.length || c.questions?.length || c.answers?.length || c.withdrawQuestionIds?.length || c.bumpQuestionIds?.length || c.questionDependencies?.length);
-var ContactWireSchema = z4.object({
-  operationId: z4.string().uuid(),
+var ContactWireSchema = z5.object({
+  operationId: z5.string().uuid(),
   ...sendFields,
-  questions: z4.array(QuestionWireSchema).min(1).max(20).optional()
+  questions: z5.array(QuestionWireSchema).min(1).max(20).optional()
 }).strict().refine(sendsAnything, "a contact sends at least one update, question, answer, withdrawal or question dependency");
 var CONTACT_SCHEMA = { type: "object", ...mcpInputSchema(ContactSchema) };
 var CONTACT_DESCRIPTION = "Send to and receive from the person: completions, updates, questions, answers to questions addressed to you, withdrawals, and what a question holds up. Every Goal named must exist: an unknown one refuses the whole contact. Each item applies on its own: read `results`, where a failed one names why.\n\n" + ASKING_RULES + `
@@ -331,13 +396,14 @@ LENGTH: the question, message or answer is the one point, at most ${ASK_MAX} cha
 UPDATES ask nothing. One reaches the person only when they asked for it, it answers something they said, or its Goal is done; else it is kept as progress. One that reaches them joins an open report card or a live call.
 
 RECEIVING: contact({}) waits ~45 s; contact({wait:false}) returns \`events\` at once. Confirm the ones you handled with contact({ackEventIds}).`;
-var GUIDE = '# Paigy\n\nThe user is remote. Reach them only through Paigy\'s tools, never the terminal. The server\'s own instructions and each tool\'s description are the contract: read them. Every read ends in `next`, the one step to take.\n\nThis same guide is inside the MCP: the `help` tool (whole, or one `topic`) and the `/mcp__paigy__help` prompt. `/mcp__paigy__welcome` starts a session.\n\nHosted (no machine): skip Pairing and Listening; answers wake you.\n\n## Pairing\n\n- Call `configure` first. With a device credential on this machine it hatches you at once.\n- Otherwise print its `user_message` to the user, then do what the result says:\n  - `awaiting_approval`: call `configure` again with its `device_code` in the same turn. On `pending`, call again.\n  - `awaiting_confirmation`: end your turn with the code as your whole reply; call `configure` with the `device_code` next turn. Calling it in the same turn drops the code.\n- When it returns `enable_command`, print it for the user to run. Do not run it yourself.\n- `configure({ logOut: true })` logs out only this agent, and only when the user asks.\n\n## Listening\n\n- Call `listen` once per session.\n  - `listening`: done.\n  - `receiving`: a listener runs but cannot wake you. Do not start a second.\n  - `unpaired`: call `configure`.\n  - `start`: run its `command` exactly, now.\n- Claude Code: run it with the **Monitor** tool, never a background Bash command. Start it again whenever the monitor expires.\n- Codex: run its `command` as a long-running background process. It queues each answer into this thread (`CODEX_THREAD_ID`).\n- With no `listen` tool, run `npx -y -p @paigy/mcp@latest paigy-listen --brief` the same way.\n- One listener per session. A `paigy-listen` in `ps` belongs to another session; ask `listen`, never `ps`.\n- Each line it prints is work: receive, read the Goal, act.\n\n## Receiving\n\n- On startup and after every wake or listener line: `contact({ wait: false })`.\n- Events: `question` (you owe an answer: send `answers`), `update` (something new on your Goal: an answer, a reply, a review), `instruction` (work sent to you).\n- Read the Goal with `get_goals` before acting. An answer arrives as an `update` event and lands on the question\'s Goal. A question reading `replied` means they chose no option: settle it from their words.\n- Acknowledge each event you handled: `contact({ ackEventIds })`. Reading acknowledges nothing; events repeat until acknowledged. A `question` event is answered, never acknowledged away.\n- `contact({ wait: true })` waits about 45 s. On `expired`, wait again without re-sending, or stop. No while-loops: while any Call is open, keep receiving through empty windows; otherwise stop at one empty window.\n\n## Asking\n\n- One question per object in `questions`, at most 80 characters. Context goes in `units`: at most 12, title \u226450, body \u2264200, one idea each.\n- `goalId` names an existing Goal. Create it with `manage_goals` first.\n- Give options when you have them: up to 6, label \u226450 naming the choice in full, `hint` \u2264200 saying what it does. Never "Other" or "Option 2". `pickMode`: `one` (default), `many`, `rank`.\n- `workItBlocks` only when you are truly stopped: it asks for a call.\n- Read the Goal before asking. Never re-ask an open or answered question, or one whose work shipped.\n- Take back a question that no longer matters: `withdrawQuestionIds`.\n- An answer you give (`answers`) is at most 80 characters.\n\n## Updates and completions\n\n- An update (`updates`: `message` \u226480, `goalIds`) asks nothing. It reaches the user only when they asked for it, when it answers something they said, or when its Goal is done; otherwise it stays on the Goal as progress.\n- Updates never complete a Goal.\n- `reply` is the button in the user\'s voice instead of "Got it", \u226450 characters ("Sounds good, I\'ll test later"). Required (`reply_required`) when the update is delivered on request (`userExplicitlyRequested`) or any Goal it names is already done.\n- Set `userExplicitlyRequested` only to relay the user\'s own words: `"any"` for "send it to me", `"call"` for "call me". Never to make a message louder.\n- Finish with a completion (`contact.completions`, or `manage_goals` kind `complete`): `goalId`, `message` \u226450, `reply` \u226450, optional `units`, optional next-step `options` {label, hint}. It closes and reports in one step. Say only what your evidence supports.\n- Read `results`: each item applies on its own, and a failed one names why.\n\n## Goals\n\n- `manage_goals` `changes`, applied in order:\n  - `create`: `title` (1-5 words) and `outcome` (one paragraph, \u2264500). More work is child Goals: `parentGoalId`. New ids come back in `results[i].goalId`.\n  - `assign`: `ownerId`, a participant as `check_activity` shows it.\n  - `defer`: `until`. You are woken when it passes. This is how you follow up later.\n  - `dependency`: `add`/`remove`; `goalId` waits on `dependsOnGoalId` to `start` or `complete`.\n  - `edit`, `move`, `state` (open, completed, canceled).\n- Any open question on a Goal, anyone\'s, blocks its completion (`goal_blocked`). Before completing, withdraw yours, finish or move its required children, and clear its dependencies.\n- On `goal_revision_conflict`, read the Goal again and reconsider.\n- Before overlapping work, read `check_activity` and the Goal\'s other contributors. Every Goal of your person is open to you; your first write puts you on it.\n- `search` finds earlier words, Goals and answers. Nothing found is not proof of absence.\n\n## Calls\n\n- Call only when the user asked: `userExplicitlyRequested: "call"` on that item. A question with `workItBlocks` may ring too.\n- Any contact while the user is on a call joins that call (`joinedCall` in `results`). Several questions arrive as one call.\n- While any Call is open, keep receiving.\n- An open Call Delivery is not proof it rang. `get_delivery` reads a Call\'s state, and with `transcript: true` what was said.\n';
-var HELP_TOPICS = ["pairing", "listening", "receiving", "asking", "updates", "goals", "calls"];
+var GUIDE = '# Paigy\n\nThe user is remote. Reach them only through Paigy\'s tools, never the terminal. The server\'s own instructions and each tool\'s description are the contract: read them. Every read ends in `next`, the one step to take.\n\nThis same guide is inside the MCP: the `help` tool (whole, or one `topic`) and the `/mcp__paigy__help` prompt. `/mcp__paigy__welcome` starts a session; `/mcp__paigy__organize` sets up a person\'s work as a board.\n\nHosted (no machine): skip Pairing and Listening; answers wake you.\n\n## Pairing\n\n- Call `configure` first. With a device credential on this machine it hatches you at once.\n- Otherwise print its `user_message` to the user, then do what the result says:\n  - `awaiting_approval`: call `configure` again with its `device_code` in the same turn. On `pending`, call again.\n  - `awaiting_confirmation`: end your turn with the code as your whole reply; call `configure` with the `device_code` next turn. Calling it in the same turn drops the code.\n- When it returns `enable_command`, print it for the user to run. Do not run it yourself.\n- `configure({ logOut: true })` logs out only this agent, and only when the user asks.\n\n## Listening\n\n- Call `listen` once per session.\n  - `listening`: done.\n  - `receiving`: a listener runs but cannot wake you. Do not start a second.\n  - `unpaired`: call `configure`.\n  - `start`: run its `command` exactly, now.\n- Claude Code: run it with the **Monitor** tool, never a background Bash command. Start it again whenever the monitor expires.\n- Codex: run its `command` as a long-running background process. It queues each answer into this thread (`CODEX_THREAD_ID`).\n- Grok: run it with the **monitor** tool, never a background shell command. Start it again whenever the monitor expires.\n- With no `listen` tool, run `npx -y -p @paigy/mcp@latest paigy-listen --brief` the same way.\n- One listener per session. A `paigy-listen` in `ps` belongs to another session; ask `listen`, never `ps`.\n- Each line it prints is work: receive, read the Goal, act.\n\n## Receiving\n\n- On startup and after every wake or listener line: `contact({ wait: false })`.\n- Events: `question` (you owe an answer: send `answers`), `update` (something new on your Goal: an answer, a reply, a review), `instruction` (work sent to you).\n- Read the Goal with `get_goals` before acting. An answer arrives as an `update` event and lands on the question\'s Goal. A question reading `replied` means they chose no option: settle it from their words.\n- Acknowledge each event you handled: `contact({ ackEventIds })`. Reading acknowledges nothing; events repeat until acknowledged. A `question` event is answered, never acknowledged away.\n- `contact({ wait: true })` waits about 45 s. On `expired`, wait again without re-sending, or stop. No while-loops: while any Call is open, keep receiving through empty windows; otherwise stop at one empty window.\n\n## Asking\n\n- One question per object in `questions`, at most 80 characters. Context goes in `units`: at most 12, title \u226450, body \u2264200, one idea each.\n- `goalId` names an existing Goal. Create it with `manage_goals` first.\n- Give options when you have them: up to 6, label \u226450 naming the choice in full, `hint` \u2264200 saying what it does. Never "Other" or "Option 2". `pickMode`: `one` (default), `many`, `rank`.\n- `workItBlocks` only when you are truly stopped: it asks for a call.\n- Read the Goal before asking. Never re-ask an open or answered question, or one whose work shipped.\n- Take back a question that no longer matters: `withdrawQuestionIds`.\n- An answer you give (`answers`) is at most 80 characters.\n\n## Updates and completions\n\n- An update (`updates`: `message` \u226480, `goalIds`) asks nothing. It reaches the user only when they asked for it, when it answers something they said, or when its Goal is done; otherwise it stays on the Goal as progress.\n- Updates never complete a Goal.\n- One load-bearing point per report: `message` is the point they must not miss, and a second point that would go wrong or stay unclear later if they missed it is its own update. `units` are context they may open, never something they must hear.\n- `reply` is the button in the user\'s voice instead of "Got it", \u226450 characters ("Sounds good, I\'ll test later"). Required (`reply_required`) when the update is delivered on request (`userExplicitlyRequested`) or any Goal it names is already done.\n- Set `userExplicitlyRequested` only to relay the user\'s own words: `"any"` for "send it to me", `"call"` for "call me". Never to make a message louder.\n- Finish with a completion (`contact.completions`, or `manage_goals` kind `complete`): `goalId`, `message` \u226450, `reply` \u226450, optional `units`, optional next-step `options` {label, hint}. It closes and reports in one step. Say only what your evidence supports.\n- Read `results`: each item applies on its own, and a failed one names why.\n\n## Goals\n\n- `manage_goals` `changes`, applied in order:\n  - `create`: `title` (1-5 words) and `outcome` (one paragraph, \u2264500). More work is child Goals: `parentGoalId`. New ids come back in `results[i].goalId`.\n  - `assign`: `ownerId`, a participant as `check_activity` shows it.\n  - `defer`: `until`. You are woken when it passes. This is how you follow up later.\n  - `dependency`: `add`/`remove`; `goalId` waits on `dependsOnGoalId` to `start` or `complete`.\n  - `edit`, `move`, `state` (open, completed, canceled).\n- Shape your work as projects with steps. Create a project under its area (`create` with `parentGoalId` = the area), then one child Goal per step; complete each step as it lands. A top-level Goal that has children is an area, not a project: never hang steps straight off a top-level Goal. The person\'s Work tab shows a project\'s progress as its steps, "3 of 5 steps done"; a project with no steps shows none. When you touch an older Goal of yours that sits flat or under a finished project, move it into that shape (`move`). A flat Goal is still valid; never reshape another agent\'s Goals.\n- Any open question on a Goal, anyone\'s, blocks its completion (`goal_blocked`). Before completing, withdraw yours, finish or move its required children, and clear its dependencies.\n- On `goal_revision_conflict`, read the Goal again and reconsider.\n- Before overlapping work, read `check_activity` and the Goal\'s other contributors. Every Goal of your person is open to you; your first write puts you on it.\n- `search` finds earlier words, Goals and answers. Nothing found is not proof of absence.\n\n## Organize\n\nWhen the person asks you to set up or organize their work ("set up my project", "help me manage all this", several products or parts to juggle), turn it into a board they can run like a team. Their Work tab shows it.\n\n- Learn the work first: read the repo and their Goals (`get_goals`, `search`). Ask only what you cannot read, at most three questions, each its own question.\n- The shape: an area per product or business (a top-level Goal); under it a project per part that moves on its own (landing page, graphics, web app, mobile app, launch); under each project its steps, three to seven, each an outcome someone can finish and check.\n- Propose before you build. Create one Goal for the setup itself, then ask one question on it: the tree in `units` (one unit per project: its title, then its steps), options "Build it" and "Change it: I\'ll say how". Create nothing else until they answer.\n- On "Build it", one `manage_goals` call per level, since a parent\'s id comes back in `results[i].goalId`: the areas, then their projects (`parentGoalId`), then every step; then a `dependency` where a project waits on another (web on the design). Never copy work that already has a Goal: yours move into the tree (`move`); another agent\'s stay where they are, named in the proposal. Then complete the setup Goal.\n- Report how to run it, and hand over one brief per project for the person to paste as that session\'s first message. Each brief names the project\'s Goal (title and id) and says: call `listen` and keep the listener running; receive with `contact({ wait: false })` at startup and after every wake; work the steps in order and complete each as it lands; report progress as updates on the project; ask anything you are blocked on through Paigy, as a question on that step\'s Goal (`workItBlocks` only when it truly stops you), never in the terminal.\n- One session per project, plus one lead session that keeps the board current. Any agent, Claude, Codex or another, can pick up any Goal; `check_activity` shows who is on what. A subagent inside a session shares that session\'s identity and listener: it may ask through Paigy, but never starts a listener of its own.\n- Keep it current as you work: complete each step as it lands and add a step when new work appears. Never reshape another agent\'s Goals.\n\n## Calls\n\n- Call only when the user asked: `userExplicitlyRequested: "call"` on that item. A question with `workItBlocks` may ring too.\n- Any contact while the user is on a call joins that call (`joinedCall` in `results`). Several questions arrive as one call.\n- While any Call is open, keep receiving.\n- An open Call Delivery is not proof it rang. `get_delivery` reads a Call\'s state, and with `transcript: true` what was said.\n';
+var SKILL_MARK = "<!-- paigy-skill";
+var HELP_TOPICS = ["pairing", "listening", "receiving", "asking", "updates", "goals", "organize", "calls"];
 var unknownTopic = (topic) => `unknown_topic: ${JSON.stringify(topic)} is not a section of the guide. Topics: ${HELP_TOPICS.join(", ")}; or omit topic for the whole guide.`;
-var HelpSchema = z5.object({
-  topic: z5.enum(HELP_TOPICS, { errorMap: (_issue, ctx) => ({ message: unknownTopic(ctx.data) }) }).optional().describe("One section of the guide. Omit it for the whole guide.")
+var HelpSchema = z6.object({
+  topic: z6.enum(HELP_TOPICS, { errorMap: (_issue, ctx) => ({ message: unknownTopic(ctx.data) }) }).optional().describe("One section of the guide. Omit it for the whole guide.")
 }).strict();
-var HELP_DESCRIPTION = "Start here: how to use Paigy, in full \u2014 pairing, listening, receiving, asking, updates and completions, Goals and calls. Pass a topic for one section. Read-only; needs no setup.";
+var HELP_DESCRIPTION = "Start here: how to use Paigy, in full \u2014 pairing, listening, receiving, asking, updates and completions, Goals, organizing work, and calls. Pass a topic for one section. Read-only; needs no setup.";
 function section(topic) {
   const parts = GUIDE.split(/^(?=## )/m);
   const found = parts.find((p) => p.slice(3).toLowerCase().startsWith(topic));
@@ -358,7 +424,8 @@ var AGENT_PROMPTS = [
     description: "How to use Paigy: the whole guide, or one topic of it.",
     arguments: [{ name: "topic", description: `One section: ${HELP_TOPICS.join(", ")}. Omit it for the whole guide.`, required: false }]
   },
-  { name: "welcome", description: "Start here: who you are on Paigy, then listen and collect your open work." }
+  { name: "welcome", description: "Start here: who you are on Paigy, then listen and collect your open work." },
+  { name: "organize", description: "Set up the person's work as a board: areas, projects and steps, proposed on a card before anything is built." }
 ];
 function promptText(name, facts) {
   if (name === "help") {
@@ -381,6 +448,11 @@ ${helpView(parsed.data).guide}`;
 ${steps.map((s, i) => `${i + 1}. ${s}`).join("\n")}
 
 The whole guide is the \`help\` tool (or /mcp__paigy__help).`;
+  }
+  if (name === "organize") {
+    return `Help me organize my work with Paigy: propose a board of areas, projects and steps, and build it once I approve. Follow this:
+
+${helpView({ topic: "organize" }).guide}`;
   }
   throw new Error(`unknown_prompt: ${JSON.stringify(name)}. Prompts: ${AGENT_PROMPTS.map((p) => p.name).join(", ")}.`);
 }
@@ -388,13 +460,13 @@ function promptResult(name, facts) {
   const text = promptText(name, facts);
   return { description: AGENT_PROMPTS.find((p) => p.name === name).description, messages: [{ role: "user", content: { type: "text", text } }] };
 }
-var CreateGoalSchema = z6.object({
-  outcome: z6.string().trim().min(1).max(1e4),
+var CreateGoalSchema = z7.object({
+  outcome: z7.string().trim().min(1).max(1e4),
   /** The work's NAME (#2115) — one to five words, how a person refers to it out loud ("the night
    *  rings"). Omit it and the brain writes one at admission from the outcome. */
-  title: z6.string().trim().min(1).max(80).optional(),
-  ownerParticipant: z6.string().trim().min(1).optional(),
-  idempotencyKey: z6.string().trim().min(1).max(200),
+  title: z7.string().trim().min(1).max(80).optional(),
+  ownerParticipant: z7.string().trim().min(1).optional(),
+  idempotencyKey: z7.string().trim().min(1).max(200),
   /** THE GOAL THIS ONE BELONGS UNDER (owner, 2026-09-15: "the ask I gave for the design doc
    *  didn't get created as a child goal of the voice UI goal, which is how it should've
    *  worked"). It could not have been: this door took no parent, so the only route was
@@ -402,63 +474,63 @@ var CreateGoalSchema = z6.object({
    *  took the short one. The hierarchy has been modelled since Goals existed and had been used
    *  ZERO times in 2,031 of them. Absent, the server judges it against the caller's open Goals
    *  (`apps/api/src/goal/intake.ts`). The writer refuses a Goal belonging to another account. */
-  parentGoalId: z6.string().uuid().optional(),
+  parentGoalId: z7.string().uuid().optional(),
   /** The repository or project identifier this Goal belongs to (#2280) — e.g. "owner/repo" or
    *  repo name. Delegated work inherits this from its parent Goal when omitted. */
-  repo: z6.string().trim().min(1).max(200).optional()
+  repo: z7.string().trim().min(1).max(200).optional()
 });
-var UpdateGoalSchema = z6.object({
-  revision: z6.number().int().positive(),
-  changes: z6.object({
-    outcome: z6.string().trim().min(1).max(1e4).optional(),
+var UpdateGoalSchema = z7.object({
+  revision: z7.number().int().positive(),
+  changes: z7.object({
+    outcome: z7.string().trim().min(1).max(1e4).optional(),
     /** The work's NAME (#2115) — one to five words, how a person refers to it out loud. The
      *  brain writes one at admission; this is the owner saying it better. Null clears it. */
-    title: z6.string().trim().min(1).max(80).nullable().optional(),
-    ownerParticipant: z6.string().trim().min(1).optional(),
-    parentGoalId: z6.string().uuid().nullable().optional(),
-    dependencies: z6.array(z6.object({ goalId: z6.string().uuid(), gate: z6.enum(["start", "finish"]) }).strict()).optional(),
-    children: z6.array(z6.object({ outcome: z6.string().trim().min(1).max(1e4), ownerParticipant: z6.string().trim().min(1), gate: z6.enum(["start", "finish"]).optional() }).strict()).optional(),
-    state: z6.enum(["active", "done", "cancelled"]).optional(),
-    progress: z6.string().trim().min(1).max(1e4).optional(),
-    reviewed: z6.literal(true).optional(),
-    dueAt: z6.string().datetime({ offset: true }).nullable().optional()
+    title: z7.string().trim().min(1).max(80).nullable().optional(),
+    ownerParticipant: z7.string().trim().min(1).optional(),
+    parentGoalId: z7.string().uuid().nullable().optional(),
+    dependencies: z7.array(z7.object({ goalId: z7.string().uuid(), gate: z7.enum(["start", "finish"]) }).strict()).optional(),
+    children: z7.array(z7.object({ outcome: z7.string().trim().min(1).max(1e4), ownerParticipant: z7.string().trim().min(1), gate: z7.enum(["start", "finish"]).optional() }).strict()).optional(),
+    state: z7.enum(["active", "done", "cancelled"]).optional(),
+    progress: z7.string().trim().min(1).max(1e4).optional(),
+    reviewed: z7.literal(true).optional(),
+    dueAt: z7.string().datetime({ offset: true }).nullable().optional()
   }).strict().refine((v) => Object.keys(v).length > 0),
-  reason: z6.string().trim().min(1).max(2e3),
-  operationId: z6.string().uuid().optional()
+  reason: z7.string().trim().min(1).max(2e3),
+  operationId: z7.string().uuid().optional()
 }).strict();
-var changedGoal = z6.string().uuid().describe("The Goal to change, as a read shows it.");
-var goalTitle = z6.string().trim().min(1).max(80).describe("The Goal's short display name: one to five words, how a person refers to it out loud.");
+var changedGoal = z7.string().uuid().describe("The Goal to change, as a read shows it.");
+var goalTitle = z7.string().trim().min(1).max(80).describe("The Goal's short display name: one to five words, how a person refers to it out loud.");
 var GOAL_OUTCOME_MAX = 500;
-var goalOutcome = z6.string().trim().min(1).max(GOAL_OUTCOME_MAX, `outcome_too_long: a Goal's outcome is one paragraph, at most ${GOAL_OUTCOME_MAX} characters; split the rest into child Goals (create with parentGoalId).`).describe(`The desired result, one paragraph of at most ${GOAL_OUTCOME_MAX} characters. More than that is more work: create child Goals under it (parentGoalId). There is no third description field.`);
-var CreateChange = z6.object({
-  kind: z6.literal("create"),
+var goalOutcome = z7.string().trim().min(1).max(GOAL_OUTCOME_MAX, `outcome_too_long: a Goal's outcome is one paragraph, at most ${GOAL_OUTCOME_MAX} characters; split the rest into child Goals (create with parentGoalId).`).describe(`The desired result, one paragraph of at most ${GOAL_OUTCOME_MAX} characters. More than that is more work: create child Goals under it (parentGoalId). There is no third description field.`);
+var CreateChange = z7.object({
+  kind: z7.literal("create"),
   title: goalTitle,
   outcome: goalOutcome,
-  ownerId: z6.string().trim().min(1).optional().describe("Who owns the work: an agent's participant (as check_activity shows it) or the person's. Omitted: you."),
-  parentGoalId: z6.string().uuid().optional().describe("The Goal it belongs under. Omitted: a root."),
-  sourceEntryIds: z6.array(z6.string().uuid()).max(20).optional().describe("The whole Entries the work came from, ones you can read.")
+  ownerId: z7.string().trim().min(1).optional().describe("Who owns the work: an agent's participant (as check_activity shows it) or the person's. Omitted: you."),
+  parentGoalId: z7.string().uuid().optional().describe("The Goal it belongs under. Omitted: a root. A project with several parts: create it, then one child per step under it; the person's Work tab counts a project's progress in those steps."),
+  sourceEntryIds: z7.array(z7.string().uuid()).max(20).optional().describe("The whole Entries the work came from, ones you can read.")
 }).strict();
-var EditChange = z6.object({ kind: z6.literal("edit"), goalId: changedGoal, title: goalTitle.optional(), outcome: goalOutcome.optional() }).strict();
-var StateChange = z6.object({ kind: z6.literal("state"), goalId: changedGoal, state: z6.enum(["open", "completed", "canceled"]).describe(
+var EditChange = z7.object({ kind: z7.literal("edit"), goalId: changedGoal, title: goalTitle.optional(), outcome: goalOutcome.optional() }).strict();
+var StateChange = z7.object({ kind: z7.literal("state"), goalId: changedGoal, state: z7.enum(["open", "completed", "canceled"]).describe(
   "completed: the Goal's own work is done now (refused while required children are open); open: reopen it; canceled: it will not be done."
 ) }).strict();
-var AssignChange = z6.object({ kind: z6.literal("assign"), goalId: changedGoal, ownerId: z6.string().trim().min(1).describe("The new owner's participant.") }).strict();
-var DeferChange = z6.object({ kind: z6.literal("defer"), goalId: changedGoal, until: z6.string().datetime({ offset: true }).nullable().describe(
+var AssignChange = z7.object({ kind: z7.literal("assign"), goalId: changedGoal, ownerId: z7.string().trim().min(1).describe("The new owner's participant.") }).strict();
+var DeferChange = z7.object({ kind: z7.literal("defer"), goalId: changedGoal, until: z7.string().datetime({ offset: true }).nullable().describe(
   "Postpone it until this instant: it waits until then, and its owner is woken when it passes (a promise to follow up later). null takes the postponement back."
 ) }).strict();
-var MoveChange = z6.object({ kind: z6.literal("move"), goalId: changedGoal, parentGoalId: z6.string().uuid().nullable().describe(
+var MoveChange = z7.object({ kind: z7.literal("move"), goalId: changedGoal, parentGoalId: z7.string().uuid().nullable().describe(
   "Its new parent, or null for a root. Only this Goal moves; its new parent's other children stay."
 ) }).strict();
-var DependencyChange = z6.object({
-  kind: z6.literal("dependency"),
-  change: z6.enum(["add", "remove"]),
+var DependencyChange = z7.object({
+  kind: z7.literal("dependency"),
+  change: z7.enum(["add", "remove"]),
   goalId: changedGoal,
-  dependsOnGoalId: z6.string().uuid().describe("The Goal it waits on."),
-  action: z6.enum(["start", "complete"]).describe("What waits: starting goalId, or completing it."),
-  reason: z6.string().trim().min(1).max(250).optional().describe("Why, at most 250 characters, as an explanation, never as policy.")
+  dependsOnGoalId: z7.string().uuid().describe("The Goal it waits on."),
+  action: z7.enum(["start", "complete"]).describe("What waits: starting goalId, or completing it."),
+  reason: z7.string().trim().min(1).max(250).optional().describe("Why, at most 250 characters, as an explanation, never as policy.")
 }).strict();
-var read = { revision: z6.number().int().positive().optional() };
-var GoalChangeSchema = z6.discriminatedUnion("kind", [
+var read = { revision: z7.number().int().positive().optional() };
+var GoalChangeSchema = z7.discriminatedUnion("kind", [
   CreateChange,
   CompleteChangeSchema.extend(read),
   EditChange.extend(read),
@@ -470,226 +542,233 @@ var GoalChangeSchema = z6.discriminatedUnion("kind", [
 ]);
 var editNames = (changes, ctx) => changes.forEach((c, i) => {
   if (c.kind === "edit" && c.title === void 0 && c.outcome === void 0) {
-    ctx.addIssue({ code: z6.ZodIssueCode.custom, path: ["changes", i], message: "an edit supplies title, outcome, or both" });
+    ctx.addIssue({ code: z7.ZodIssueCode.custom, path: ["changes", i], message: "an edit supplies title, outcome, or both" });
   }
 });
-var ManageGoalsSchema = z6.object({
-  requestId: z6.string().uuid().optional(),
-  changes: z6.array(GoalChangeSchema).min(1).max(50)
+var ManageGoalsSchema = z7.object({
+  requestId: z7.string().uuid().optional(),
+  changes: z7.array(GoalChangeSchema).min(1).max(50)
 }).strict().superRefine((v, ctx) => editNames(v.changes, ctx));
-var ManageGoalsToolSchema = z6.object({
-  changes: z6.array(z6.discriminatedUnion("kind", [CreateChange, CompleteChangeSchema, EditChange, StateChange, AssignChange, DeferChange, MoveChange, DependencyChange])).min(1).max(50).describe("The changes, applied in order, each on its own.")
+var ManageGoalsToolSchema = z7.object({
+  changes: z7.array(z7.discriminatedUnion("kind", [CreateChange, CompleteChangeSchema, EditChange, StateChange, AssignChange, DeferChange, MoveChange, DependencyChange])).min(1).max(50).describe("The changes, applied in order, each on its own.")
 }).strict().superRefine((v, ctx) => editNames(v.changes, ctx));
-var MANAGE_GOALS_DESCRIPTION = "Create, edit, assign, organize or close work. Every change here changes Goals; answering, withdrawing or asking Questions stays in contact. Each change applies independently: a refused one names why (`results[i].error`, e.g. goal_revision_conflict, goal_children_open, goal_not_found) and the rest still apply, so read every result: a partial result is never a complete success. Kinds: complete (goalId, message capped at 50 characters, units, reply, optional next-step options {label, hint}) closes and reports atomically, exactly as contact.completions; create (title, outcome, ownerId, parentGoalId, sourceEntryIds) returns the new Goal's id in `results[i].goalId`, in the order requested, to use in later calls (an outcome is one paragraph, at most 500 characters: split more into child Goals with parentGoalId); edit (title and/or outcome); state (open, completed, canceled); assign (ownerId); defer (until: the Goal waits until then and its owner is woken when it passes; null takes it back); move (parentGoalId, or null for a root: only this Goal moves); dependency (add or remove: goalId waits on dependsOnGoalId to start or to complete). A Goal cannot be completed while its required children or dependencies remain open: finish or move them first. Finishing the children does not prove the parent's own work is done. An edit applies at the version you last read: if someone changed the Goal since, it is refused as goal_revision_conflict, so read it again (get_goals) and reconsider. You may change any Goal of your person; a change puts you on it. Returns `ok` (every change applied), `results` per change (applied or failed), and `goals`, each changed Goal's id, state, title, outcome and revision.";
-var GetGoalsSchema = z6.object({
-  goalIds: z6.array(z6.string().uuid()).min(1).max(10).describe("The Goals to read, in order; one to ten IDs."),
+var MANAGE_GOALS_DESCRIPTION = "Create, edit, assign, organize or close work. SHAPE IT AS PROJECTS WITH STEPS: create a project under its area (parentGoalId; a top-level Goal with children is an area), then each part as its child Goal, so the person sees N of M steps. Move a flat or orphaned Goal of yours into that shape when you touch it; a flat Goal is valid. Every change here changes Goals; answering, withdrawing or asking Questions stays in contact. Each change applies independently: a refused one names why (`results[i].error`, e.g. goal_revision_conflict, goal_children_open, goal_not_found) and the rest still apply, so read every result: a partial result is never a complete success. Kinds: complete (goalId, message capped at 50 characters, units, reply, optional next-step options {label, hint}) closes and reports atomically, exactly as contact.completions; create (title, outcome, ownerId, parentGoalId, sourceEntryIds) returns the new Goal's id in `results[i].goalId`, in the order requested, to use in later calls (an outcome is one paragraph, at most 500 characters: split more into child Goals with parentGoalId); edit (title and/or outcome); state (open, completed, canceled); assign (ownerId); defer (until: the Goal waits until then and its owner is woken when it passes; null takes it back); move (parentGoalId, or null for a root: only this Goal moves); dependency (add or remove: goalId waits on dependsOnGoalId to start or to complete). A Goal cannot be completed while its required children or dependencies remain open: finish or move them first. Finishing the children does not prove the parent's own work is done. An edit applies at the version you last read: if someone changed the Goal since, it is refused as goal_revision_conflict, so read it again (get_goals) and reconsider. You may change any Goal of your person; a change puts you on it. Returns `ok` (every change applied), `results` per change (applied or failed), and `goals`, each changed Goal's id, state, title, outcome and revision.";
+var GetGoalsSchema = z7.object({
+  goalIds: z7.array(z7.string().uuid()).min(1).max(10).describe("The Goals to read, in order; one to ten IDs."),
   /** Every entry in full. Without it the read carries the person's words, open questions and your
    *  newest entry, and counts what it left out (`apps/api/src/goal/collapse.ts`). */
-  history: z6.boolean().optional()
+  history: z7.boolean().optional()
 }).strict();
-var GetDeliverySchema = z6.object({
-  deliveryId: z6.string().uuid().describe("The Delivery id returned by contact or receive."),
+var GetDeliverySchema = z7.object({
+  deliveryId: z7.string().uuid().describe("The Delivery id returned by contact or receive."),
   // WHAT WAS SAID ON A CALL (owner, 2026-10-26, Goal c743f4a4: "A flag on get_delivery", "Any of my agents").
-  transcript: z6.boolean().optional().describe("For a Call: also return what was said on it, one page of spoken lines."),
-  after: z6.string().max(40).optional().describe("With transcript: the next value the previous page returned, to read the next page.")
+  transcript: z7.boolean().optional().describe("For a Call: also return what was said on it, one page of spoken lines."),
+  after: z7.string().max(40).optional().describe("With transcript: the next value the previous page returned, to read the next page.")
 }).strict();
 var GET_DELIVERY_DESCRIPTION = "Read a Delivery in full: its whole ask, units, options and answers, plus current callState and callOutcome for a Call. For a Call, transcript: true adds what was said: each spoken line in order, who said it, and when from pickup, a page at a time. A line from the room, not the person, is marked room. Any agent of the same account may read it; another account is not disclosed. Reading never acknowledges or resends anything. Use this after a call or when receive names a Call you are waiting on.";
 var GET_GOALS_DESCRIPTION = "Read one to ten Goals by goalIds, returned in the same order under goals (a read never puts you on them). history applies to every requested Goal. A refused read fails the call with its original refusal; no partial result is presented as complete. Each Goal includes its ten most recent other contributors (names, latest entry headlines and times), outcome, state, revision, progress, blockers, the conversation on it (each question with its options and what was decided), `next`, the one step to take, and `more`, where to read further. A question in that conversation reads `open` (nothing yet), `answered` (a choice was made, and `answer` carries it), `replied` (they said something and the read settled the question on their words \u2014 NO option of yours was chosen, and the words are the reply line beside it, so read that before you act), or `closed`. The conversation carries everything the person said, every open question and your newest entry; your older entries are left out and counted \u2014 history: true reads every entry in full. Every Goal of your person is readable, whichever of their agents owns it; another account's Goals are not disclosed.";
-var SearchToolSchema = z6.object({
-  query: z6.string().trim().min(1).max(500),
-  types: z6.array(z6.enum(["entry", "goal", "answer"])).min(1).optional(),
-  goalId: z6.string().uuid().optional(),
-  limit: z6.number().int().min(1).max(20).optional()
+var SearchToolSchema = z7.object({
+  query: z7.string().trim().min(1).max(500),
+  types: z7.array(z7.enum(["entry", "goal", "answer"])).min(1).optional(),
+  goalId: z7.string().uuid().optional(),
+  limit: z7.number().int().min(1).max(20).optional()
 }).strict();
 var SEARCH_DESCRIPTION = "Search your person's history across all of their agents: Entries (what anyone said or wrote, typed or spoken on a call), Goals (by title and outcome) and replies -- the person's answers, acknowledgements and deferrals (type answer; found by their summary, by the question or report they reply to, or by the words that gave them). query is words to look for; records sharing more of its words rank first, and exact names work. types narrows it to entry, goal and/or answer (default: all three). goalId searches under one Goal: its Entries, the replies to its items, and it and its immediate children. limit is matches per type, 1 to 20 (default 8). Read-only. Each match carries its whole saved words, its ID and its links (a reply carries its kind -- answered, acknowledged or deferred -- whether it is closed (a deferral is not), its summary in under ten words, its question or report, the choice made and the Entries that support it); `omitted` counts what matched but was left out, so narrow the words or add a goalId to see it. Nothing found is not proof that nothing exists; a refused search says why. Sealed (encrypted) content is never searched or returned.";
-var ConfigureIdentitySchema = z6.object({}).strict();
+var READS = { readOnlyHint: true, destructiveHint: false, openWorldHint: false };
+var SENDS = { readOnlyHint: false, destructiveHint: true, openWorldHint: false };
+var ConfigureIdentitySchema = z7.object({}).strict();
 var CONFIGURE_IDENTITY_DESCRIPTION = "Read this authenticated agent\u2019s current name, voice and token ID. Read-only; no arguments. Pairing, renaming and logout are available only through the local configure driver.";
-var CheckActivitySchema = z6.object({}).strict();
+var CheckActivitySchema = z7.object({}).strict();
 var FEEDBACK_TEXT_MAX = 5e4;
-var SendFeedbackSchema = z6.object({
-  text: z6.string().trim().min(1).max(FEEDBACK_TEXT_MAX).describe(`The report, in markdown: what happened, what was expected, how to reproduce it, versions, evidence. At most ${FEEDBACK_TEXT_MAX} characters.`),
-  title: z6.string().trim().min(1).max(200).optional().describe("One line naming it."),
-  kind: z6.enum(["bug", "idea", "other"]).default("other").describe("bug: something Paigy does wrong. idea: something it should do. other: anything else.")
+var SendFeedbackSchema = z7.object({
+  text: z7.string().trim().min(1).max(FEEDBACK_TEXT_MAX).describe(`The report, in markdown: what happened, what was expected, how to reproduce it, versions, evidence. At most ${FEEDBACK_TEXT_MAX} characters.`),
+  title: z7.string().trim().min(1).max(200).optional().describe("One line naming it."),
+  kind: z7.enum(["bug", "idea", "other"]).default("other").describe("bug: something Paigy does wrong. idea: something it should do. other: anything else.")
 }).strict();
 var SEND_FEEDBACK_DESCRIPTION = "Send feedback about Paigy itself (a bug report, an idea) straight to the Paigy team. Send it only when the person asked you to, or agreed when you offered: it goes to Paigy, never to the person, and nobody answers it here. Not for your work or a question for the person: use contact for those. Returns whether it was stored.";
 var AGENT_TOOLS = [
-  { name: "configure", description: CONFIGURE_IDENTITY_DESCRIPTION, inputSchema: mcpInputSchema(ConfigureIdentitySchema) },
-  { name: "contact", description: CONTACT_DESCRIPTION, inputSchema: CONTACT_SCHEMA },
-  { name: "manage_goals", description: MANAGE_GOALS_DESCRIPTION, inputSchema: mcpInputSchema(ManageGoalsToolSchema) },
-  { name: "get_delivery", description: GET_DELIVERY_DESCRIPTION, inputSchema: mcpInputSchema(GetDeliverySchema) },
-  { name: "get_goals", description: GET_GOALS_DESCRIPTION, inputSchema: mcpInputSchema(GetGoalsSchema) },
-  { name: "search", description: SEARCH_DESCRIPTION, inputSchema: mcpInputSchema(SearchToolSchema) },
-  { name: "check_activity", description: "Read your person's agents that contributed in the last 24 hours, newest first: at most 30 agents and their five most recently touched Goals, with their latest entry headline and time. Use get_goals to read a Goal and its other contributors before starting overlapping work. Read-only; no arguments.", inputSchema: mcpInputSchema(CheckActivitySchema) },
-  { name: "send_feedback", description: SEND_FEEDBACK_DESCRIPTION, inputSchema: mcpInputSchema(SendFeedbackSchema) },
+  { name: "configure", title: "Who am I", annotations: READS, description: CONFIGURE_IDENTITY_DESCRIPTION, inputSchema: mcpInputSchema(ConfigureIdentitySchema) },
+  { name: "contact", title: "Contact the person", annotations: SENDS, description: CONTACT_DESCRIPTION, inputSchema: CONTACT_SCHEMA },
+  { name: "manage_goals", title: "Manage Goals", annotations: SENDS, description: MANAGE_GOALS_DESCRIPTION, inputSchema: mcpInputSchema(ManageGoalsToolSchema) },
+  { name: "get_delivery", title: "Read a delivery", annotations: READS, description: GET_DELIVERY_DESCRIPTION, inputSchema: mcpInputSchema(GetDeliverySchema) },
+  { name: "get_goals", title: "Read Goals", annotations: READS, description: GET_GOALS_DESCRIPTION, inputSchema: mcpInputSchema(GetGoalsSchema) },
+  { name: "search", title: "Search history", annotations: READS, description: SEARCH_DESCRIPTION, inputSchema: mcpInputSchema(SearchToolSchema) },
+  { name: "check_activity", title: "Recent agent activity", annotations: READS, description: "Read your person's agents that contributed in the last 24 hours, newest first: at most 30 agents and their five most recently touched Goals, with their latest entry headline and time. Use get_goals to read a Goal and its other contributors before starting overlapping work. Read-only; no arguments.", inputSchema: mcpInputSchema(CheckActivitySchema) },
+  { name: "send_feedback", title: "Send feedback to Paigy", annotations: SENDS, description: SEND_FEEDBACK_DESCRIPTION, inputSchema: mcpInputSchema(SendFeedbackSchema) },
   // THE GUIDE FROM INSIDE THE MCP (owner, 2026-10-27): the skill's text, needing no identity (`help.ts`).
-  { name: "help", description: HELP_DESCRIPTION, inputSchema: mcpInputSchema(HelpSchema) }
+  { name: "help", title: "Paigy guide", annotations: READS, description: HELP_DESCRIPTION, inputSchema: mcpInputSchema(HelpSchema) }
 ];
 var AGENT_TOOL_NAMES = AGENT_TOOLS.map((t) => t.name);
 var LISTEN_COMMAND = "npx -y -p @paigy/mcp@latest paigy-listen --brief";
-var LISTEN_MONITOR = "your harness's MONITOR tool, the kind that wakes you on every line it prints (in Claude Code: Monitor, not a background Bash command, which only tells you when it exits)";
+var LISTEN_MONITOR = "your harness's MONITOR tool, the kind that wakes you on every line it prints (in Claude Code: Monitor, not a background Bash command, which only tells you when it exits; in Grok: the monitor tool, the same way)";
 function serverInstructions(opts) {
   const waits = opts.waits ? `contact({wait:true}) waits ~45 s for a reply to what you sent; contact({}) for anything addressed to you. Call \`listen\` once to wake at rest.` : "A contact here returns after one read; answers wake you, and contact({wait:false}) collects them.";
-  return `New here? Call help. On startup and after a wake, contact({wait:false}) returns your events. They repeat until acknowledged with contact({ackEventIds}); reading does not. get_goals reads a Goal's conversation; your first write puts you on it, and every Goal of your person is open to you. manage_goals creates, assigns and organizes work; to follow up, defer the Goal with manage_goals (woken when due). Complete and report at once: contact.completions or manage_goals kind:complete. Every read ends in \`next\`: your next step. ${waits} Report progress as a contact update; one they don't need yet stays on the Goal. An open Call is not proof it rang. Your user is remote: reach them only through Paigy, never the terminal.
+  return `New here? Call help. On startup and after a wake, contact({wait:false}) returns your events. They repeat until acknowledged with contact({ackEventIds}); reading doesn't. get_goals reads a Goal; your first write puts you on it; every Goal of your person is open to you. manage_goals creates, assigns and organizes work as projects with steps; to follow up, defer it (woken when due). Complete and report at once: contact.completions or manage_goals kind:complete. Every read ends in \`next\`: your next step. ${waits} Report progress as a contact update (what changed for them; causedBy, shipped); one they don't need yet stays on the Goal. An open Call isn't proof it rang. Your user is remote: reach them only through Paigy, never the terminal.
 
 ${ASKING_RULES}`;
 }
-var id = z7.string().uuid();
-var key = z7.string().min(1).max(40);
-var participant = z7.string().regex(/^(human|agent):.+$/, "a participant ID such as agent:<uuid>");
-var SourceSchema = z7.object({ entryId: id }).strict();
-var ResultRefSchema = z7.union([z7.object({ id }).strict(), z7.object({ local: key }).strict()]);
-var NodeRefSchema = z7.object({ type: z7.enum(["goal", "question"]), id }).strict();
-var DependencySchema = z7.object({
+var SKILL = '---\nname: paigy\ndescription: Reach the user through Paigy. Use whenever you need to tell the user something, ask them a question, get a decision or approval, call their phone, wait for or receive their answers, start or check the Paigy listener, or create, assign, defer or complete Paigy Goals.\n---\n<!-- paigy-skill \xB7 source: paigy monorepo packs/skill/paigy/SKILL.md; mirrored to paigy-ai/claude plugins/paigy/skills/paigy/SKILL.md and written by https://paigy.ai/install, which replaces only a file carrying this line. Edit the source, never a copy. -->\n\n# Paigy\n\nThe user is remote. Reach them only through Paigy\'s tools, never the terminal. The server\'s own instructions and each tool\'s description are the contract: read them. Every read ends in `next`, the one step to take.\n\nThis same guide is inside the MCP: the `help` tool (whole, or one `topic`) and the `/mcp__paigy__help` prompt. `/mcp__paigy__welcome` starts a session; `/mcp__paigy__organize` sets up a person\'s work as a board.\n\nHosted (no machine): skip Pairing and Listening; answers wake you.\n\n## Pairing\n\n- Call `configure` first. With a device credential on this machine it hatches you at once.\n- Otherwise print its `user_message` to the user, then do what the result says:\n  - `awaiting_approval`: call `configure` again with its `device_code` in the same turn. On `pending`, call again.\n  - `awaiting_confirmation`: end your turn with the code as your whole reply; call `configure` with the `device_code` next turn. Calling it in the same turn drops the code.\n- When it returns `enable_command`, print it for the user to run. Do not run it yourself.\n- `configure({ logOut: true })` logs out only this agent, and only when the user asks.\n\n## Listening\n\n- Call `listen` once per session.\n  - `listening`: done.\n  - `receiving`: a listener runs but cannot wake you. Do not start a second.\n  - `unpaired`: call `configure`.\n  - `start`: run its `command` exactly, now.\n- Claude Code: run it with the **Monitor** tool, never a background Bash command. Start it again whenever the monitor expires.\n- Codex: run its `command` as a long-running background process. It queues each answer into this thread (`CODEX_THREAD_ID`).\n- Grok: run it with the **monitor** tool, never a background shell command. Start it again whenever the monitor expires.\n- With no `listen` tool, run `npx -y -p @paigy/mcp@latest paigy-listen --brief` the same way.\n- One listener per session. A `paigy-listen` in `ps` belongs to another session; ask `listen`, never `ps`.\n- Each line it prints is work: receive, read the Goal, act.\n\n## Receiving\n\n- On startup and after every wake or listener line: `contact({ wait: false })`.\n- Events: `question` (you owe an answer: send `answers`), `update` (something new on your Goal: an answer, a reply, a review), `instruction` (work sent to you).\n- Read the Goal with `get_goals` before acting. An answer arrives as an `update` event and lands on the question\'s Goal. A question reading `replied` means they chose no option: settle it from their words.\n- Acknowledge each event you handled: `contact({ ackEventIds })`. Reading acknowledges nothing; events repeat until acknowledged. A `question` event is answered, never acknowledged away.\n- `contact({ wait: true })` waits about 45 s. On `expired`, wait again without re-sending, or stop. No while-loops: while any Call is open, keep receiving through empty windows; otherwise stop at one empty window.\n\n## Asking\n\n- One question per object in `questions`, at most 80 characters. Context goes in `units`: at most 12, title \u226450, body \u2264200, one idea each.\n- `goalId` names an existing Goal. Create it with `manage_goals` first.\n- Give options when you have them: up to 6, label \u226450 naming the choice in full, `hint` \u2264200 saying what it does. Never "Other" or "Option 2". `pickMode`: `one` (default), `many`, `rank`.\n- `workItBlocks` only when you are truly stopped: it asks for a call.\n- Read the Goal before asking. Never re-ask an open or answered question, or one whose work shipped.\n- Take back a question that no longer matters: `withdrawQuestionIds`.\n- An answer you give (`answers`) is at most 80 characters.\n\n## Updates and completions\n\n- An update (`updates`: `message` \u226480, `goalIds`) asks nothing. It reaches the user only when they asked for it, when it answers something they said, or when its Goal is done; otherwise it stays on the Goal as progress.\n- Updates never complete a Goal.\n- One load-bearing point per report: `message` is the point they must not miss, and a second point that would go wrong or stay unclear later if they missed it is its own update. `units` are context they may open, never something they must hear.\n- `reply` is the button in the user\'s voice instead of "Got it", \u226450 characters ("Sounds good, I\'ll test later"). Required (`reply_required`) when the update is delivered on request (`userExplicitlyRequested`) or any Goal it names is already done.\n- Set `userExplicitlyRequested` only to relay the user\'s own words: `"any"` for "send it to me", `"call"` for "call me". Never to make a message louder.\n- Finish with a completion (`contact.completions`, or `manage_goals` kind `complete`): `goalId`, `message` \u226450, `reply` \u226450, optional `units`, optional next-step `options` {label, hint}. It closes and reports in one step. Say only what your evidence supports.\n- Read `results`: each item applies on its own, and a failed one names why.\n\n## Goals\n\n- `manage_goals` `changes`, applied in order:\n  - `create`: `title` (1-5 words) and `outcome` (one paragraph, \u2264500). More work is child Goals: `parentGoalId`. New ids come back in `results[i].goalId`.\n  - `assign`: `ownerId`, a participant as `check_activity` shows it.\n  - `defer`: `until`. You are woken when it passes. This is how you follow up later.\n  - `dependency`: `add`/`remove`; `goalId` waits on `dependsOnGoalId` to `start` or `complete`.\n  - `edit`, `move`, `state` (open, completed, canceled).\n- Shape your work as projects with steps. Create a project under its area (`create` with `parentGoalId` = the area), then one child Goal per step; complete each step as it lands. A top-level Goal that has children is an area, not a project: never hang steps straight off a top-level Goal. The person\'s Work tab shows a project\'s progress as its steps, "3 of 5 steps done"; a project with no steps shows none. When you touch an older Goal of yours that sits flat or under a finished project, move it into that shape (`move`). A flat Goal is still valid; never reshape another agent\'s Goals.\n- Any open question on a Goal, anyone\'s, blocks its completion (`goal_blocked`). Before completing, withdraw yours, finish or move its required children, and clear its dependencies.\n- On `goal_revision_conflict`, read the Goal again and reconsider.\n- Before overlapping work, read `check_activity` and the Goal\'s other contributors. Every Goal of your person is open to you; your first write puts you on it.\n- `search` finds earlier words, Goals and answers. Nothing found is not proof of absence.\n\n## Organize\n\nWhen the person asks you to set up or organize their work ("set up my project", "help me manage all this", several products or parts to juggle), turn it into a board they can run like a team. Their Work tab shows it.\n\n- Learn the work first: read the repo and their Goals (`get_goals`, `search`). Ask only what you cannot read, at most three questions, each its own question.\n- The shape: an area per product or business (a top-level Goal); under it a project per part that moves on its own (landing page, graphics, web app, mobile app, launch); under each project its steps, three to seven, each an outcome someone can finish and check.\n- Propose before you build. Create one Goal for the setup itself, then ask one question on it: the tree in `units` (one unit per project: its title, then its steps), options "Build it" and "Change it: I\'ll say how". Create nothing else until they answer.\n- On "Build it", one `manage_goals` call per level, since a parent\'s id comes back in `results[i].goalId`: the areas, then their projects (`parentGoalId`), then every step; then a `dependency` where a project waits on another (web on the design). Never copy work that already has a Goal: yours move into the tree (`move`); another agent\'s stay where they are, named in the proposal. Then complete the setup Goal.\n- Report how to run it, and hand over one brief per project for the person to paste as that session\'s first message. Each brief names the project\'s Goal (title and id) and says: call `listen` and keep the listener running; receive with `contact({ wait: false })` at startup and after every wake; work the steps in order and complete each as it lands; report progress as updates on the project; ask anything you are blocked on through Paigy, as a question on that step\'s Goal (`workItBlocks` only when it truly stops you), never in the terminal.\n- One session per project, plus one lead session that keeps the board current. Any agent, Claude, Codex or another, can pick up any Goal; `check_activity` shows who is on what. A subagent inside a session shares that session\'s identity and listener: it may ask through Paigy, but never starts a listener of its own.\n- Keep it current as you work: complete each step as it lands and add a step when new work appears. Never reshape another agent\'s Goals.\n\n## Calls\n\n- Call only when the user asked: `userExplicitlyRequested: "call"` on that item. A question with `workItBlocks` may ring too.\n- Any contact while the user is on a call joins that call (`joinedCall` in `results`). Several questions arrive as one call.\n- While any Call is open, keep receiving.\n- An open Call Delivery is not proof it rang. `get_delivery` reads a Call\'s state, and with `transcript: true` what was said.\n';
+var id = z8.string().uuid();
+var key = z8.string().min(1).max(40);
+var participant = z8.string().regex(/^(human|agent):.+$/, "a participant ID such as agent:<uuid>");
+var SourceSchema = z8.object({ entryId: id }).strict();
+var ResultRefSchema = z8.union([z8.object({ id }).strict(), z8.object({ local: key }).strict()]);
+var NodeRefSchema = z8.object({ type: z8.enum(["goal", "question"]), id }).strict();
+var DependencySchema = z8.object({
   blocker: NodeRefSchema,
   blocked: NodeRefSchema,
-  action: z7.enum(["start", "complete", "answer"]),
-  reason: z7.string().max(500).optional()
+  action: z8.enum(["start", "complete", "answer"]),
+  reason: z8.string().max(500).optional()
 }).strict();
-var goalState = z7.enum(["open", "completed", "canceled"]);
+var goalState = z8.enum(["open", "completed", "canceled"]);
 var existingGoalChanges = [
-  z7.object({ kind: z7.literal("edit"), goalId: id, title: z7.string().min(1).max(120).optional(), outcome: z7.string().min(1).optional() }).strict().refine((c) => c.title !== void 0 || c.outcome !== void 0, "an edit changes the title or the outcome"),
-  z7.object({ kind: z7.literal("state"), goalId: id, state: goalState }).strict(),
-  z7.object({ kind: z7.literal("assign"), goalId: id, ownerId: participant }).strict(),
-  z7.object({ kind: z7.literal("defer"), goalId: id, until: z7.string().datetime().nullable() }).strict(),
-  z7.object({ kind: z7.literal("move"), goalId: id, parentGoalId: id.nullable() }).strict(),
-  z7.object({
-    kind: z7.literal("dependency"),
-    change: z7.enum(["add", "remove"]),
+  z8.object({ kind: z8.literal("edit"), goalId: id, title: z8.string().min(1).max(120).optional(), outcome: z8.string().min(1).optional() }).strict().refine((c) => c.title !== void 0 || c.outcome !== void 0, "an edit changes the title or the outcome"),
+  z8.object({ kind: z8.literal("state"), goalId: id, state: goalState }).strict(),
+  z8.object({ kind: z8.literal("assign"), goalId: id, ownerId: participant }).strict(),
+  z8.object({ kind: z8.literal("defer"), goalId: id, until: z8.string().datetime().nullable() }).strict(),
+  z8.object({ kind: z8.literal("move"), goalId: id, parentGoalId: id.nullable() }).strict(),
+  z8.object({
+    kind: z8.literal("dependency"),
+    change: z8.enum(["add", "remove"]),
     goalId: id,
     dependsOnGoalId: id,
-    action: z7.enum(["start", "complete"]),
-    reason: z7.string().max(500).optional()
+    action: z8.enum(["start", "complete"]),
+    reason: z8.string().max(500).optional()
   }).strict()
 ];
-var BrainGoalChangeSchema = z7.union([
+var BrainGoalChangeSchema = z8.union([
   // newSession: Paigy starts a session to own it (owner, 2026-10-06); until it has, the person owns it.
-  z7.object({
-    kind: z7.literal("create"),
-    outcome: z7.string().min(1),
-    title: z7.string().min(1).max(120),
+  z8.object({
+    kind: z8.literal("create"),
+    outcome: z8.string().min(1),
+    title: z8.string().min(1).max(120),
     ownerId: participant,
     parentGoal: ResultRefSchema.nullable(),
-    newSession: z7.literal(true).optional()
+    newSession: z8.literal(true).optional()
   }).strict(),
   ...existingGoalChanges
 ]);
-var option = z7.object({ id: z7.string().min(1).max(40), label: z7.string().min(1), hint: z7.string().optional() }).strict();
-var QuestionChangeSchema = z7.union([
-  z7.object({
-    kind: z7.literal("create"),
-    text: z7.string().min(1),
-    summary: z7.string().optional(),
+var option = z8.object({ id: z8.string().min(1).max(40), label: z8.string().min(1), hint: z8.string().optional() }).strict();
+var QuestionChangeSchema = z8.union([
+  z8.object({
+    kind: z8.literal("create"),
+    text: z8.string().min(1),
+    summary: z8.string().optional(),
     answererId: participant,
     goal: ResultRefSchema.optional(),
-    blocks: z7.array(z7.object({ question: ResultRefSchema }).strict()),
-    options: z7.array(option).min(2).optional(),
-    pickMode: z7.enum(["one", "many", "rank"]).optional()
+    blocks: z8.array(z8.object({ question: ResultRefSchema }).strict()),
+    options: z8.array(option).min(2).optional(),
+    pickMode: z8.enum(["one", "many", "rank"]).optional()
   }).strict().refine((c) => c.options === void 0 === (c.pickMode === void 0), "options and pickMode come together"),
-  z7.object({ kind: z7.literal("edit"), questionId: id, text: z7.string().min(1) }).strict(),
-  z7.object({ kind: z7.literal("assign"), questionId: id, answererId: participant }).strict(),
-  z7.object({ kind: z7.literal("withdraw"), questionId: id }).strict(),
-  z7.object({ kind: z7.literal("dependency"), change: z7.enum(["add", "remove"]), edge: DependencySchema }).strict()
+  z8.object({ kind: z8.literal("edit"), questionId: id, text: z8.string().min(1) }).strict(),
+  z8.object({ kind: z8.literal("assign"), questionId: id, answererId: participant }).strict(),
+  z8.object({ kind: z8.literal("withdraw"), questionId: id }).strict(),
+  z8.object({ kind: z8.literal("dependency"), change: z8.enum(["add", "remove"]), edge: DependencySchema }).strict()
 ]);
-var lessonScope = z7.union([
-  z7.object({ kind: z7.literal("user") }).strict(),
-  z7.object({ kind: z7.literal("goal"), goal: ResultRefSchema }).strict()
+var lessonScope = z8.union([
+  z8.object({ kind: z8.literal("user") }).strict(),
+  z8.object({ kind: z8.literal("goal"), goal: ResultRefSchema }).strict()
 ]);
-var LessonChangeSchema = z7.union([
-  z7.object({ kind: z7.literal("remember"), text: z7.string().min(1), scope: lessonScope }).strict(),
-  z7.object({ kind: z7.literal("revise"), lessonId: id, text: z7.string().min(1), scope: lessonScope }).strict(),
-  z7.object({ kind: z7.literal("withdraw"), lessonId: id }).strict()
+var LessonChangeSchema = z8.union([
+  z8.object({ kind: z8.literal("remember"), text: z8.string().min(1), scope: lessonScope }).strict(),
+  z8.object({ kind: z8.literal("revise"), lessonId: id, text: z8.string().min(1), scope: lessonScope }).strict(),
+  z8.object({ kind: z8.literal("withdraw"), lessonId: id }).strict()
 ]);
-var BrainNextSchema = z7.object({
-  instructions: z7.array(z7.object({ entryKeys: z7.array(key).min(1) }).strict()),
-  say: z7.array(key),
-  then: z7.enum(["listen", "hold", "end", "none"]),
-  waitFor: z7.array(ResultRefSchema),
-  reason: z7.string()
+var BrainNextSchema = z8.object({
+  instructions: z8.array(z8.object({ entryKeys: z8.array(key).min(1) }).strict()),
+  say: z8.array(key),
+  then: z8.enum(["listen", "hold", "end", "none"]),
+  waitFor: z8.array(ResultRefSchema),
+  reason: z8.string()
 }).strict();
 var messageBase = {
   key,
-  text: z7.string().min(1),
-  goals: z7.array(ResultRefSchema),
-  questions: z7.array(ResultRefSchema),
-  entryKeys: z7.array(key),
-  obligationIds: z7.array(id)
+  text: z8.string().min(1),
+  goals: z8.array(ResultRefSchema),
+  questions: z8.array(ResultRefSchema),
+  entryKeys: z8.array(key),
+  obligationIds: z8.array(id)
 };
-var BrainMessageSchema = z7.union([
+var BrainMessageSchema = z8.union([
   // A MESSAGE TO THE PERSON KEEPS ITS SENTENCES (turns.md): the talker writes them one by one, and each is
   // its own unit of speech (`<cycle>:<key>:<n>`). `text` is the same words joined, for every reader of a
   // message's words. Absent on results written before sentences.
-  z7.object({
+  z8.object({
     ...messageBase,
-    to: z7.object({ kind: z7.literal("user") }).strict(),
-    sentences: z7.array(z7.string().min(1)).min(1).optional(),
+    to: z8.object({ kind: z8.literal("user") }).strict(),
+    sentences: z8.array(z8.string().min(1)).min(1).optional(),
     // THE REPORTS IT TELLS, BY THEIR OWN IDENTITY (call 9d1c096c, 2026-10-15): the talker's message names the report
     // Entries it speaks of, so the plan's line names that report rather than re-finding it by its Goal. Absent on
     // results written before it.
-    tells: z7.array(z7.object({ id }).strict()).optional()
+    tells: z8.array(z8.object({ id }).strict()).optional()
   }).strict(),
-  z7.object({
+  z8.object({
     ...messageBase,
-    to: z7.object({ kind: z7.literal("agent"), agentId: z7.string().regex(/^agent:(?!paigy$)\S+$/) }).strict(),
-    call: z7.object({ callId: id, listen: z7.literal(true) }).strict().optional(),
+    to: z8.object({ kind: z8.literal("agent"), agentId: z8.string().regex(/^agent:(?!paigy$)\S+$/) }).strict(),
+    call: z8.object({ callId: id, listen: z8.literal(true) }).strict().optional(),
     // WHAT IT ASKS THE AGENT, IN UNDER TEN WORDS (owner, 2026-10-15: "Yes, add the summary"): the talker's own, stored
     // as the sent Entry's summary, which the recap's "Sent to <agent>: …" row shows. Absent on results written before it.
-    summary: z7.string().optional(),
+    summary: z8.string().optional(),
+    // THE PERSON'S LINES IT CARRIES, IN THEIR WORDS (owner, 2026-11-05: "just deliver my exact words to the agent"): the
+    // `entries` keys of those lines, each relayed whole on the agent's Delivery (`context = false`), where the rest of
+    // `entryKeys` ride as context. Absent on results written before it, and on a message that carries none.
+    carries: z8.array(key).optional(),
     // THE PERSON ASKED TO BE CALLED BACK ONCE THIS AGENT'S WORK IS DONE (turns.md, step 0): the agent owes
     // the call, and its next contact about it rings them (`public.callbacks`).
-    callback: z7.literal(true).optional()
+    callback: z8.literal(true).optional()
   }).strict()
 ]);
-var BrainSearchSchema = z7.object({
-  query: z7.string().min(1).max(500),
-  within: z7.array(z7.enum(["entries", "goals", "answers"])).min(1),
+var BrainSearchSchema = z8.object({
+  query: z8.string().min(1).max(500),
+  within: z8.array(z8.enum(["entries", "goals", "answers"])).min(1),
   goalId: id.optional()
 }).strict();
-var BrainResultSchema = z7.object({
-  messages: z7.array(BrainMessageSchema),
-  entries: z7.array(z7.object({ key, sources: z7.array(SourceSchema).min(1), goals: z7.array(ResultRefSchema), summary: z7.string().optional() }).strict()),
-  questions: z7.array(z7.object({ key, entryKeys: z7.array(key), change: QuestionChangeSchema }).strict()),
-  answers: z7.array(z7.object({
+var BrainResultSchema = z8.object({
+  messages: z8.array(BrainMessageSchema),
+  entries: z8.array(z8.object({ key, sources: z8.array(SourceSchema).min(1), goals: z8.array(ResultRefSchema), summary: z8.string().optional() }).strict()),
+  questions: z8.array(z8.object({ key, entryKeys: z8.array(key), change: QuestionChangeSchema }).strict()),
+  answers: z8.array(z8.object({
     question: ResultRefSchema,
-    entryKeys: z7.array(key).min(1),
-    summary: z7.string(),
-    selectedOptionIds: z7.array(z7.string().min(1)).optional()
+    entryKeys: z8.array(key).min(1),
+    summary: z8.string(),
+    selectedOptionIds: z8.array(z8.string().min(1)).optional()
   }).strict()),
-  goals: z7.array(z7.object({ key, entryKeys: z7.array(key), change: BrainGoalChangeSchema }).strict()),
-  feedback: z7.array(z7.object({ entryKeys: z7.array(key).min(1) }).strict()),
-  lessons: z7.array(z7.object({ entryKeys: z7.array(key).min(1), change: LessonChangeSchema }).strict()),
+  goals: z8.array(z8.object({ key, entryKeys: z8.array(key), change: BrainGoalChangeSchema }).strict()),
+  feedback: z8.array(z8.object({ entryKeys: z8.array(key).min(1) }).strict()),
+  lessons: z8.array(z8.object({ entryKeys: z8.array(key).min(1), change: LessonChangeSchema }).strict()),
   next: BrainNextSchema,
   search: BrainSearchSchema.optional()
 }).strict();
-var str2 = z7.string();
-var strs = z7.array(z7.string());
-var CompactResultSchema = z7.object({
-  next: z7.object({ say: strs, then: z7.enum(["listen", "hold", "end", "none"]), waitFor: strs, reason: str2, instructions: strs }).strict(),
-  messages: z7.array(z7.object({ key: str2, to: str2, text: str2, about: strs, entries: strs, owed: strs, inviteToCall: z7.boolean() }).strict()),
-  entries: z7.array(z7.object({ key: str2, from: str2, on: strs }).strict()),
-  answers: z7.array(z7.object({ question: str2, entries: strs, summary: str2, picked: strs }).strict()),
-  changes: z7.array(z7.object({
+var str2 = z8.string();
+var strs = z8.array(z8.string());
+var CompactResultSchema = z8.object({
+  next: z8.object({ say: strs, then: z8.enum(["listen", "hold", "end", "none"]), waitFor: strs, reason: str2, instructions: strs }).strict(),
+  messages: z8.array(z8.object({ key: str2, to: str2, text: str2, about: strs, entries: strs, owed: strs, inviteToCall: z8.boolean() }).strict()),
+  entries: z8.array(z8.object({ key: str2, from: str2, on: strs }).strict()),
+  answers: z8.array(z8.object({ question: str2, entries: strs, summary: str2, picked: strs }).strict()),
+  changes: z8.array(z8.object({
     key: str2,
-    what: z7.enum(["goal", "question", "lesson"]),
-    op: z7.enum(["create", "edit", "assign", "open", "complete", "cancel", "defer", "move", "block", "unblock", "withdraw", "remember", "revise"]),
+    what: z8.enum(["goal", "question", "lesson"]),
+    op: z8.enum(["create", "edit", "assign", "open", "complete", "cancel", "defer", "move", "block", "unblock", "withdraw", "remember", "revise"]),
     id: str2,
     text: str2,
     title: str2,
     who: str2,
     under: str2,
-    gate: z7.enum(["start", "complete", "answer", "none"]),
-    options: z7.array(z7.object({ id: str2, label: str2 }).strict()),
-    pick: z7.enum(["one", "many", "rank", "words"]),
+    gate: z8.enum(["start", "complete", "answer", "none"]),
+    options: z8.array(z8.object({ id: str2, label: str2 }).strict()),
+    pick: z8.enum(["one", "many", "rank", "words"]),
     entries: strs
   }).strict()),
-  feedback: z7.array(strs),
-  search: z7.object({ query: str2, within: strs, goal: str2 }).strict()
+  feedback: z8.array(strs),
+  search: z8.object({ query: str2, within: strs, goal: str2 }).strict()
 }).strict();
 var COMPACT_RESULT_JSON_SCHEMA = (({ $schema: _, ...rest }) => rest)(
   zodToJsonSchema2(CompactResultSchema, { $refStrategy: "none" })
 );
-var str3 = z8.string();
-var strs2 = z8.array(z8.string());
-var TalkerReplySchema = z8.object({
+var str3 = z9.string();
+var strs2 = z9.array(z9.string());
+var TalkerReplySchema = z9.object({
   /** What to say and send, messages to the person first: spoken in the order written. `to` is "user"
    *  or an agent's handle; `sentences` is what a message to the person says, one short sentence each
    *  ([] to an agent); `text` is what a message to an agent says ("" to the person); `summary` is what it
@@ -697,37 +776,45 @@ var TalkerReplySchema = z8.object({
    *  "Yes, add the summary"), which the recap's "Sent to <agent>: …" row shows; `about` names
    *  handles; `blocks` is the item a question to an agent must be answered before ("" for none);
    *  `callback` is true on a message to the agent doing work the person asked to be called back about
-   *  once it is done (false otherwise): that agent then owes the call (turns.md, step 0). */
-  messages: z8.array(z8.object({ key: str3, to: str3, sentences: strs2, text: str3, summary: str3, about: strs2, blocks: str3, callback: z8.boolean() }).strict()),
+   *  once it is done (false otherwise): that agent then owes the call (turns.md, step 0). `lines` are the
+   *  person's line handles a message to an agent carries ([] for none): the agent receives those lines, in their
+   *  words, and `text` is then only what the talker adds beyond them (owner, 2026-11-05: "just deliver my exact
+   *  words to the agent, keeping the summary as a way for me to navigate them more easily"). */
+  messages: z9.array(z9.object({ key: str3, to: str3, sentences: strs2, text: str3, summary: str3, about: strs2, blocks: str3, callback: z9.boolean(), lines: strs2.optional() }).strict()),
   /** What the call does next: listen, hold (the person asked for a moment) or end (they asked to). */
-  then: z8.enum(["listen", "hold", "end"]),
+  then: z9.enum(["listen", "hold", "end"]),
   /** The person's replies to items (replies, owner 2026-10-07): the item's handle, the chosen option
    *  IDs, the line handles, how they replied in under ten words (keeping any condition), and `defer`:
    *  "" when their words settle the item, else what a reply that puts it off waits for -- "call" (the
    *  end of this call), a question's handle (that question's answer) or an ISO time. */
-  replies: z8.array(z8.object({ item: str3, options: strs2, lines: strs2, summary: str3, defer: str3 }).strict()),
+  replies: z9.array(z9.object({ item: str3, options: strs2, lines: strs2, summary: str3, defer: str3 }).strict()),
   /** The line handles that tell the talker how to run this call; they last until it ends and reach no agent. */
   instruction: strs2,
   /** The line handles the filer acts on, and the only ones it reads (3e): new work, a change to existing
    *  work, feedback on Paigy, anything said to an agent, and any rule meant to outlast the call. */
   file: strs2,
   /** Evidence to fetch for a second round: a search, an item's full text, or an agent by name. */
-  need: z8.object({ kind: z8.enum(["search", "item", "agent", ""]), text: str3 }).strict(),
+  need: z9.object({ kind: z9.enum(["search", "item", "agent", ""]), text: str3 }).strict(),
   /** The handle of the item to explain ("" for none). */
   explain: str3
 }).strict();
-var TALKER_REPLY_JSON_SCHEMA = (({ $schema: _, ...rest }) => rest)(
+var TALKER_REPLY_JSON_SCHEMA = withEveryMessageField((({ $schema: _, ...rest }) => rest)(
   zodToJsonSchema3(TalkerReplySchema, { $refStrategy: "none" })
-);
-var str4 = z9.string();
-var strs3 = z9.array(z9.string());
-var FilerReplySchema = z9.object({
+));
+function withEveryMessageField(schema) {
+  const item = schema.properties?.messages?.items;
+  if (item?.properties) item.required = Object.keys(item.properties);
+  return schema;
+}
+var str4 = z10.string();
+var strs3 = z10.array(z10.string());
+var FilerReplySchema = z10.object({
   /** Which Goals each new line goes on: a line's handle, Goal handles (or keys of new Goals: a line that asks
    *  for new work goes on the Goal it creates, owner 2026-10-26), and what the line says in at most twenty
    *  words (owner, 2026-10-26: "bump it from 10 to 20 words"). */
-  filed: z9.array(z9.object({ line: str4, goals: strs3, summary: str4 }).strict()),
-  goals: z9.array(z9.object({
-    op: z9.enum(["create", "edit", "complete", "cancel", "reopen", "assign", "defer", "move", "block", "unblock"]),
+  filed: z10.array(z10.object({ line: str4, goals: strs3, summary: str4 }).strict()),
+  goals: z10.array(z10.object({
+    op: z10.enum(["create", "edit", "complete", "cancel", "reopen", "assign", "defer", "move", "block", "unblock"]),
     key: str4,
     goal: str4,
     title: str4,
@@ -736,33 +823,33 @@ var FilerReplySchema = z9.object({
     under: str4,
     until: str4,
     on: str4,
-    gate: z9.enum(["start", "complete", "answer", ""]),
+    gate: z10.enum(["start", "complete", "answer", ""]),
     why: str4
   }).strict()),
-  questions: z9.array(z9.object({
-    op: z9.enum(["create", "edit", "assign", "withdraw"]),
+  questions: z10.array(z10.object({
+    op: z10.enum(["create", "edit", "assign", "withdraw"]),
     key: str4,
     question: str4,
     text: str4,
     summary: str4,
     to: str4,
     goal: str4,
-    options: z9.array(z9.object({ id: str4, label: str4 }).strict()),
-    pick: z9.enum(["one", "many", "rank", "words", ""]),
+    options: z10.array(z10.object({ id: str4, label: str4 }).strict()),
+    pick: z10.enum(["one", "many", "rank", "words", ""]),
     blocks: str4
   }).strict()),
-  lessons: z9.array(z9.object({ op: z9.enum(["remember", "revise", "withdraw"]), lesson: str4, text: str4, scope: str4 }).strict()),
+  lessons: z10.array(z10.object({ op: z10.enum(["remember", "revise", "withdraw"]), lesson: str4, text: str4, scope: str4 }).strict()),
   /** Line handles whose words comment on Paigy itself. */
   feedback: strs3
 }).strict();
 var FILER_REPLY_JSON_SCHEMA = (({ $schema: _, ...rest }) => rest)(
   zodToJsonSchema4(FilerReplySchema, { $refStrategy: "none" })
 );
-var ExplainerReplySchema = z10.object({
+var ExplainerReplySchema = z11.object({
   /** The explanation, as sentences, spoken in order. */
-  say: z10.array(z10.string()),
+  say: z11.array(z11.string()),
   /** Questions to an agent about gaps in what it read: the agent's handle and the question. */
-  followUps: z10.array(z10.object({ to: z10.string(), text: z10.string() }).strict())
+  followUps: z11.array(z11.object({ to: z11.string(), text: z11.string() }).strict())
 }).strict();
 var EXPLAINER_REPLY_JSON_SCHEMA = (({ $schema: _, ...rest }) => rest)(
   zodToJsonSchema5(ExplainerReplySchema, { $refStrategy: "none" })
@@ -785,49 +872,49 @@ var DISTILLER_DEFECTS = [
   "dropped"
   // something it had to say never played and was never said later (owner, 2026-10-08, call 65433a9d)
 ];
-var DistillerReplySchema = z11.object({
-  defects: z11.array(z11.object({
-    kind: z11.enum(DISTILLER_DEFECTS),
+var DistillerReplySchema = z12.object({
+  defects: z12.array(z12.object({
+    kind: z12.enum(DISTILLER_DEFECTS),
     /** The handles of the lines it shows in. */
-    lines: z11.array(z11.string()),
+    lines: z12.array(z12.string()),
     /** The handle of the record it concerns, or "". */
-    about: z11.string(),
+    about: z12.string(),
     /** One plain sentence, in words, never handles: it is what the feedback row keeps. */
-    why: z11.string()
+    why: z12.string()
   }).strict())
 }).strict();
 var DISTILLER_REPLY_JSON_SCHEMA = (({ $schema: _, ...rest }) => rest)(
   zodToJsonSchema6(DistillerReplySchema, { $refStrategy: "none" })
 );
 var TIDY_AUDIENCES = ["talker", "filer", "agents"];
-var TidyReplySchema = z11.object({
-  lessons: z11.array(z11.object({
-    lesson: z11.string(),
-    op: z11.enum(["keep", "revise", "withdraw"]),
-    text: z11.string(),
-    for: z11.enum([...TIDY_AUDIENCES, ""]),
-    into: z11.string()
+var TidyReplySchema = z12.object({
+  lessons: z12.array(z12.object({
+    lesson: z12.string(),
+    op: z12.enum(["keep", "revise", "withdraw"]),
+    text: z12.string(),
+    for: z12.enum([...TIDY_AUDIENCES, ""]),
+    into: z12.string()
   }).strict())
 }).strict();
 var TIDY_REPLY_JSON_SCHEMA = (({ $schema: _, ...rest }) => rest)(
   zodToJsonSchema6(TidyReplySchema, { $refStrategy: "none" })
 );
-var HandoutReplySchema = z12.object({
-  assign: z12.array(z12.object({
+var HandoutReplySchema = z13.object({
+  assign: z13.array(z13.object({
     /** The Goal's handle (G). */
-    goal: z12.string(),
+    goal: z13.string(),
     /** The agent's handle (A). */
-    to: z12.string(),
+    to: z13.string(),
     /** At most ten words, naming the Goal of theirs it is most like: the reason the hand-out records. */
-    why: z12.string()
+    why: z13.string()
   }).strict()),
   /** The Goals no listed agent's work is even loosely like: Paigy starts a session for them (owner, 2026-10-24,
    *  Goal 37603006, "One per family of related Goals"). */
-  new: z12.array(z12.object({
+  new: z13.array(z13.object({
     /** The Goal's handle (G). */
-    goal: z12.string(),
+    goal: z13.string(),
     /** At most ten words: why nothing active fits it. */
-    why: z12.string()
+    why: z13.string()
   }).strict())
 }).strict();
 var HANDOUT_REPLY_JSON_SCHEMA = (({ $schema: _, ...rest }) => rest)(
@@ -861,20 +948,20 @@ var LIVE_MS = 3 * 6e4;
 var WORKING_MS = 60 * 6e4;
 var WORKING_TIER_MS = 2 * 60 * 6e4;
 var RECENT_TIER_MS = 24 * 60 * 6e4;
-var ContextSchema = z13.object({
-  title: z13.string().min(1).describe("One-line headline of what you need (required, non-empty)."),
-  description: z13.array(z13.string().min(1)).describe(
+var ContextSchema = z14.object({
+  title: z14.string().min(1).describe("One-line headline of what you need (required, non-empty)."),
+  description: z14.array(z14.string().min(1)).describe(
     "Semantic chunks of detail (each a standalone, non-empty piece). The user can select chunks to ask you to expand. MAY BE EMPTY: a claim whose whole content is its heading \u2014 a single sentence \u2014 has no body, and saying so beats repeating the heading underneath itself. That repeat is what `min(1)` used to force, at 2x the storage, with every reader subtracting it back out at render time."
   ),
   /** THE ASK'S UNITS (3c): its context as titled units, in order. The app lists the titles and opens
    *  a body on a tap. Absent on an ask sent without units, which renders as before. */
-  units: z13.array(z13.object({ title: z13.string(), body: z13.string() })).optional()
+  units: z14.array(z14.object({ title: z14.string(), body: z14.string() })).optional()
 });
-var ParticipantSchema = z13.object({
-  kind: z13.enum(["human", "agent"]),
-  id: z13.string()
+var ParticipantSchema = z14.object({
+  kind: z14.enum(["human", "agent"]),
+  id: z14.string()
 });
-var TransformSchema = z13.enum([
+var TransformSchema = z14.enum([
   "structure",
   // shape intent into an answer contract; pick channel/urgency — broker `ask`, `select` shapes, `points`
   "request_more",
@@ -890,13 +977,13 @@ var TransformSchema = z13.enum([
   "summarize"
   // reduce volume, keep decision value — 30-turn cap, spoken briefing
 ]);
-var VisualSchema = z13.object({
-  url: z13.string().url(),
-  label: z13.string().optional()
+var VisualSchema = z14.object({
+  url: z14.string().url(),
+  label: z14.string().optional()
 });
-var NotifyLevelSchema = z13.enum(["inbox", "push", "banner", "call"]);
-var SelectShapeSchema = z13.enum(["one", "many", "rank", "confirm", "text"]);
-var ReceiptEventSchema = z13.enum([
+var NotifyLevelSchema = z14.enum(["inbox", "push", "banner", "call"]);
+var SelectShapeSchema = z14.enum(["one", "many", "rank", "confirm", "text"]);
+var ReceiptEventSchema = z14.enum([
   "delivered",
   // the bundle reached the recipient at some level
   "seen",
@@ -926,47 +1013,47 @@ var ReceiptEventSchema = z13.enum([
   // be rewound by a writer that forgot to advance it.
   "restarted"
 ]);
-var AttentionSchema = z13.object({
+var AttentionSchema = z14.object({
   urgency: NotifyLevelSchema,
   /** The required answer shape, or null for a plain notify that asks nothing back. */
   select: SelectShapeSchema.nullable(),
   /** Coverage contract (#396) — points the answer must address; null = none declared. */
-  points: z13.array(z13.string()).nullable(),
+  points: z14.array(z14.string()).nullable(),
   /** Whether the ask blocks the sender — what lets arbitration escalate it on silence. */
-  blocking: z13.boolean(),
+  blocking: z14.boolean(),
   /** Reserved (docs/model/model.md lists it): a response deadline. No row column yet — a later Phase 2
    *  slice wires it; optional so today's rows/callers project cleanly. */
-  deadline: z13.string().datetime().nullable().optional()
+  deadline: z14.string().datetime().nullable().optional()
 });
-var NotifyRequestFields = z13.object({
+var NotifyRequestFields = z14.object({
   /** Plaintext message content. Present on the plaintext path (today's shape);
    *  ABSENT on the E2EE path, where the sealed `envelope` below carries it. The
    *  superRefine at the bottom enforces exactly one of the two. */
   context: ContextSchema.optional(),
-  options: z13.array(OptionSchema.omit({ id: true }).extend({ label: z13.string().trim().min(1).max(1e3) }).strict()).min(OPTIONS_MIN).max(OPTIONS_MAX).optional().describe(
+  options: z14.array(OptionSchema.omit({ id: true }).extend({ label: z14.string().trim().min(1).max(1e3) }).strict()).min(OPTIONS_MIN).max(OPTIONS_MAX).optional().describe(
     "The choices, in order \u2014 required when select is 'one'/'many'/'rank', omitted otherwise. Ids are assigned automatically by position ('1', '2', \u2026); the user's answer references them as optionId(s)."
   ),
-  points: z13.array(z13.string().min(1)).optional().describe(
+  points: z14.array(z14.string().min(1)).optional().describe(
     "The distinct things you need answered, each a short phrase \u2014 on a call the broker keeps the conversation going until each is addressed, and the reply reports which were covered, so a half-answer is never silently returned as final. Omit for single-part asks."
   ),
-  visuals: z13.array(VisualSchema).optional().describe(
+  visuals: z14.array(VisualSchema).optional().describe(
     "Images attached to the message itself \u2014 context for the whole question (a screenshot, a chart). For a preview on one selectable choice, use that option's `html`/`image` instead."
   ),
   /** Git repo the agent is working in ("owner/name"). Local MCP fills this from the checkout — omit unless overriding. */
-  repo: z13.string().optional(),
+  repo: z14.string().optional(),
   /** Git branch the agent is on. Local MCP fills this from the checkout — omit unless overriding. */
-  branch: z13.string().optional(),
+  branch: z14.string().optional(),
   /** Continue an existing conversation — the id of any notification in it (its root
    *  is the conversation's identity). Omitted = start a new conversation. Renamed
    *  from `parentId` (2026-08-03): one linkage system, the parent; the API edge
    *  still accepts the old name from older clients. */
-  parentId: z13.string().uuid().optional(),
+  parentId: z14.string().uuid().optional(),
   /** The durable outcome this contact advances. Optional during the notification-to-Work
    *  migration; when present, a blocking ask creates a DecisionNeed for this Work. */
-  workId: z13.string().uuid().optional(),
+  workId: z14.string().uuid().optional(),
   /** Target Goal scope. During staged migration this is accepted by the shared contract but
    * target delivery activation remains model-gated; workId and goalId are mutually exclusive. */
-  goalId: z13.string().uuid().optional(),
+  goalId: z14.string().uuid().optional(),
   urgency: NotifyLevelSchema.default("inbox").describe(
     "The level you're requesting \u2014 the user's account permissions + session mode can lower it. 'inbox' (default) = sits silently in the inbox for the user to get to. 'push' = a quiet passive push (lands in Notification Center, no sound) \u2014 a gentle heads-up. 'banner' = a time-sensitive banner/lock-screen push with sound (a 'paige') they tap to open \u2014 use when you need them soon-ish but it's not worth ringing them. 'call' = rings the user's phone now (a CallKit voice call) \u2014 use only when you genuinely need them in the moment (blocked and waiting, time-sensitive). context.title is what they see on the banner/ring, so make it specific."
   ),
@@ -974,7 +1061,7 @@ var NotifyRequestFields = z13.object({
    *  visible and marks it needs_input. Renamed from the old `parentId` (2026-08-03)
    *  when `parentId` became the conversation handle: `parentId` says WHERE, this
    *  says HOW. */
-  clarifies: z13.string().optional(),
+  clarifies: z14.string().optional(),
   select: SelectShapeSchema.optional().describe(
     "How the user answers \u2014 required on the fully-shaped form, pick the shape that fits the question: 'one' = pick one option, 'many' = pick several, 'rank' = pick & order (each needs `options`); 'confirm' = yes/no or approve/deny; 'text' = free-form reply only (status updates, open questions). 'confirm' and 'text' take no options. Omit only when sending the simplified `ask` form \u2014 the broker picks the shape."
   ),
@@ -988,20 +1075,20 @@ var NotifyRequestFields = z13.object({
   // (docs/history/agenda-design.md) — not by a wire cap the agent has to pre-summarize under.
   // Owner, 2026-07-28: "our actual limitation on how long something is to the user should
   // come from the broker splitting and summarizing." The cap that remains is a size guard.
-  ask: z13.string().min(1).max(1e4).optional().describe(
+  ask: z14.string().min(1).max(1e4).optional().describe(
     'SIMPLIFIED FORM \u2014 state in plain prose what you need to learn from the user and why it matters now (e.g. "I need to know whether to deploy the auth fix \u2014 tests are green, staging verified"). Write as much as the situation needs (up to 10k characters) \u2014 Paigy breaks it into topics and reads it back a few sentences at a time; do NOT pre-summarize it into one line. Paigy derives the title, answer shape, options, and delivery channel for you. Mutually exclusive with context/select/options \u2014 send one form or the other.'
   ),
-  needs: z13.array(z13.string().min(1)).optional().describe(
+  needs: z14.array(z14.string().min(1)).optional().describe(
     "With `ask` only: the distinct things you need answered when the ask is multi-part \u2014 becomes the coverage contract (`points`), so a half-answer is never silently final."
   ),
-  urgencyHint: z13.enum(["whenever", "soon", "now"]).optional().describe(
+  urgencyHint: z14.enum(["whenever", "soon", "now"]).optional().describe(
     "With `ask` only: how urgently you need the answer \u2014 'whenever' (inbox), 'soon' (worth a heads-up), 'now' (you're blocked this minute). A hint, not a command: the user's settings still have the final word."
   ),
   /** #575: the ONE self-report that replaces urgencyHint + blocking — what happens
    *  to the agent's work while it waits. Normalized server-side into those two
    *  fields (normalizeWaiting) so everything downstream is untouched; explicit
    *  urgencyHint/blocking win when both are sent. */
-  waiting: z13.enum(["none", "soft", "hard"]).optional().describe(
+  waiting: z14.enum(["none", "soft", "hard"]).optional().describe(
     "With `ask`: what happens to your work while you wait. 'none' = you're just informing the user. 'soft' = you'd like an answer but can keep working. 'hard' = you are stopped until they answer (reaches them urgently and escalates to a real phone call if unanswered). Replaces urgencyHint + blocking \u2014 send this one field."
   ),
   /** Δ9b (#895): HOLD this claim so the sender can correct the plan before anyone is
@@ -1009,98 +1096,98 @@ var NotifyRequestFields = z13.object({
    *  holding by default would charge every quiet claim that minute before any agent could
    *  correct anything. Ignored for `waiting: 'hard'`: a blocking ask rings on what we have,
    *  and the enrichment can still land mid-call (#781 re-plans the unspoken tail). */
-  confirm: z13.boolean().optional().describe(
+  confirm: z14.boolean().optional().describe(
     "Hold this one so you can correct the plan before the user is interrupted. The response comes back with `held: true` and the plan; POST the confirm route to release it (with options/visuals/urgency corrections, or nothing at all). If you never do, it is announced anyway a couple of minutes later. Ignored when waiting is 'hard'."
   ),
   /** #575: a RELAY of the user's explicitly stated preference, never the agent's
    *  choice. Outranks waiting in both directions: 'call' rings even for a
    *  waiting:'none' "call me when it's done"; 'message' never rings even for
    *  waiting:'hard'. */
-  channel: z13.enum(["call", "message"]).optional().describe(
+  channel: z14.enum(["call", "message"]).optional().describe(
     "Only if the user explicitly said how to reach them \u2014 'call me' \u2192 'call', 'just message/text me' \u2192 'message'. Omit otherwise; Paigy picks."
   ),
-  confirmStyle: z13.enum(["yesno", "approve"]).default("yesno").describe(
+  confirmStyle: z14.enum(["yesno", "approve"]).default("yesno").describe(
     "Labels for a select:'confirm' paige \u2014 'yesno' (Yes/No) or 'approve' (Approve/Deny). Ignored unless select is 'confirm'."
   ),
-  blocking: z13.boolean().default(false).describe(
+  blocking: z14.boolean().default(false).describe(
     "Set true when real downstream work is stuck behind this specific decision \u2014 you can't make meaningful progress until it's answered. This is the real signal for how urgently the user should be reached; it's what the premier use case (an agent that stays unblocked instead of going idle) depends on. Independent of `urgency`: a `banner`-level question can still be `blocking` (something IS stuck, just not time-critical enough to ring for immediately) \u2014 if it goes unanswered a while, Paigy escalates it to a real call using this flag rather than guessing from how many other things happen to be pending. Leave false for anything you could work around, defer, or where other useful work exists meanwhile."
   )
 });
 var NotifyRequestSchema = NotifyRequestFields.superRefine((r, ctx) => {
-  if (r.workId && r.goalId) ctx.addIssue({ code: z13.ZodIssueCode.custom, path: ["goalId"], message: "pass goalId or workId, not both" });
+  if (r.workId && r.goalId) ctx.addIssue({ code: z14.ZodIssueCode.custom, path: ["goalId"], message: "pass goalId or workId, not both" });
   if (r.ask !== void 0) {
     for (const f of ["context", "select", "points"]) {
       if (r[f] !== void 0)
-        ctx.addIssue({ code: z13.ZodIssueCode.custom, path: [f], message: `the simplified \`ask\` form takes no ${f} \u2014 the broker derives the answer shape from your prose. Drop ${f} and say it in \`ask\` instead ("should I\u2026" for approve/deny, "which of these\u2026" for a pick), passing \`options\` when you're offering concrete alternatives.` });
+        ctx.addIssue({ code: z14.ZodIssueCode.custom, path: [f], message: `the simplified \`ask\` form takes no ${f} \u2014 the broker derives the answer shape from your prose. Drop ${f} and say it in \`ask\` instead ("should I\u2026" for approve/deny, "which of these\u2026" for a pick), passing \`options\` when you're offering concrete alternatives.` });
     }
     return;
   }
   if (r.needs !== void 0 || r.urgencyHint !== void 0)
-    ctx.addIssue({ code: z13.ZodIssueCode.custom, path: ["needs"], message: "needs/urgencyHint belong to the simplified `ask` form \u2014 with a shaped request use points/urgency" });
+    ctx.addIssue({ code: z14.ZodIssueCode.custom, path: ["needs"], message: "needs/urgencyHint belong to the simplified `ask` form \u2014 with a shaped request use points/urgency" });
   if (!r.context)
-    ctx.addIssue({ code: z13.ZodIssueCode.custom, path: ["context"], message: "context is required (plaintext path)" });
+    ctx.addIssue({ code: z14.ZodIssueCode.custom, path: ["context"], message: "context is required (plaintext path)" });
   if (!r.select)
-    ctx.addIssue({ code: z13.ZodIssueCode.custom, path: ["select"], message: "select is required on the shaped form" });
+    ctx.addIssue({ code: z14.ZodIssueCode.custom, path: ["select"], message: "select is required on the shaped form" });
   const needsOptions = r.select === "one" || r.select === "many" || r.select === "rank";
   if (needsOptions && !r.options?.length)
-    ctx.addIssue({ code: z13.ZodIssueCode.custom, path: ["options"], message: `select:'${r.select}' requires options` });
+    ctx.addIssue({ code: z14.ZodIssueCode.custom, path: ["options"], message: `select:'${r.select}' requires options` });
   if (!needsOptions && r.options?.length)
-    ctx.addIssue({ code: z13.ZodIssueCode.custom, path: ["options"], message: `select:'${r.select}' takes no options` });
+    ctx.addIssue({ code: z14.ZodIssueCode.custom, path: ["options"], message: `select:'${r.select}' takes no options` });
 });
-var NotifyStatusSchema = z13.enum(["pending", "answered", "ignored"]);
-var AgentStateSchema = z13.enum(["idle", "in_progress", "completed", "needs_input"]);
-var TurnSchema = z13.object({
-  prompt: z13.string(),
-  reply: z13.string()
+var NotifyStatusSchema = z14.enum(["pending", "answered", "ignored"]);
+var AgentStateSchema = z14.enum(["idle", "in_progress", "completed", "needs_input"]);
+var TurnSchema = z14.object({
+  prompt: z14.string(),
+  reply: z14.string()
 });
-var UserAnswerSchema = z13.discriminatedUnion("kind", [
-  z13.object({ kind: z13.literal("option"), optionId: z13.string(), label: z13.string().optional() }),
-  z13.object({ kind: z13.literal("text"), text: z13.string() }),
-  z13.object({ kind: z13.literal("ignored") }),
-  z13.object({ kind: z13.literal("multi"), optionIds: z13.array(z13.string()), labels: z13.array(z13.string()).optional() }),
-  z13.object({ kind: z13.literal("ranked"), optionIds: z13.array(z13.string()), labels: z13.array(z13.string()).optional() }),
-  z13.object({ kind: z13.literal("clarify"), chunks: z13.array(z13.string()).min(1) }),
-  z13.object({ kind: z13.literal("confirm"), approved: z13.boolean() }),
-  z13.object({ kind: z13.literal("turns"), turns: z13.array(TurnSchema).min(1) }),
+var UserAnswerSchema = z14.discriminatedUnion("kind", [
+  z14.object({ kind: z14.literal("option"), optionId: z14.string(), label: z14.string().optional() }),
+  z14.object({ kind: z14.literal("text"), text: z14.string() }),
+  z14.object({ kind: z14.literal("ignored") }),
+  z14.object({ kind: z14.literal("multi"), optionIds: z14.array(z14.string()), labels: z14.array(z14.string()).optional() }),
+  z14.object({ kind: z14.literal("ranked"), optionIds: z14.array(z14.string()), labels: z14.array(z14.string()).optional() }),
+  z14.object({ kind: z14.literal("clarify"), chunks: z14.array(z14.string()).min(1) }),
+  z14.object({ kind: z14.literal("confirm"), approved: z14.boolean() }),
+  z14.object({ kind: z14.literal("turns"), turns: z14.array(TurnSchema).min(1) }),
   /** An auto-answer derived from the user's PAST decisions (docs/history/precedent-design.md §2):
    *  delivered through the same settle/await path as a human answer, carrying the judge's
    *  derivation and the precedent ids it grew from. Always paired with a visible trail
    *  card the user can reply to — the broker never overrides the user. */
-  z13.object({ kind: z13.literal("precedent"), answer: z13.string(), derivation: z13.string(), sources: z13.array(z13.string()).min(1) })
+  z14.object({ kind: z14.literal("precedent"), answer: z14.string(), derivation: z14.string(), sources: z14.array(z14.string()).min(1) })
 ]);
-var IntentSchema = z13.object({
+var IntentSchema = z14.object({
   // The full vocabulary the bot's mapper emits (mapper.INTENT_KINDS) — the schema lagged
   // it by two ("detail", "feedback"), and because the settle handler parsed the array
   // all-or-nothing, ONE feedback act silently dropped EVERY intent on the call,
   // questions included. Found auditing five calls' stored feedback, 2026-08-01.
-  kind: z13.enum(["defer", "delegate", "channel", "question", "detail", "feedback", "command", "control"]),
-  detail: z13.string(),
+  kind: z14.enum(["defer", "delegate", "channel", "question", "detail", "feedback", "command", "control"]),
+  detail: z14.string(),
   /** Defer only: seconds until the callback the caller asked for, when something upstream
    *  already read the time. Nothing sets it today (#397 documented an MCP parser that was
    *  never written) — the API reads the defer's `detail` itself with `notes/when.ts`
    *  (`parseDelay`, #1292), and a value here simply wins over that reading. */
-  dueInSeconds: z13.number().int().positive().optional(),
+  dueInSeconds: z14.number().int().positive().optional(),
   /** Feedback only (#812): WHICH failure the complaint names — typed by the mapper that
    *  already read the utterance, so `signals.kind` stops defaulting to
    *  'other' on every row. A table that records that something was wrong and nothing
    *  about what cannot answer "is the bot looping less this week?". */
-  fault: z13.enum(["loop", "unanswered", "overridden", "misheard", "slow", "other"]).optional()
+  fault: z14.enum(["loop", "unanswered", "overridden", "misheard", "slow", "other"]).optional()
 });
-var RideAlongSchema = z13.object({
+var RideAlongSchema = z14.object({
   /** The note this came from — assign/clarify/close it through /api/notes/:id. */
-  noteId: z13.string(),
+  noteId: z14.string(),
   /** What to do, in the owner's own words (the note's headline). Never model-rewritten. */
-  text: z13.string(),
+  text: z14.string(),
   /** The thread to report back on, when the note was dispatched over the request rail. */
-  parentId: z13.string().nullable()
+  parentId: z14.string().nullable()
 });
-var AwaitItemSchema = z13.discriminatedUnion("type", [
-  z13.object({
-    type: z13.literal("reply"),
-    parentId: z13.string(),
-    notificationId: z13.string(),
-    workId: z13.string().uuid().optional(),
-    decisionId: z13.string().uuid().optional(),
+var AwaitItemSchema = z14.discriminatedUnion("type", [
+  z14.object({
+    type: z14.literal("reply"),
+    parentId: z14.string(),
+    notificationId: z14.string(),
+    workId: z14.string().uuid().optional(),
+    decisionId: z14.string().uuid().optional(),
     answer: UserAnswerSchema,
     /** WHAT THE AGENT CANNOT KNOW FROM THE FIELDS BESIDE IT (owner, 2026-09-04, issue
      *  #1537). One line, built from the record: the ask and the caller's reply VERBATIM,
@@ -1110,109 +1197,109 @@ var AwaitItemSchema = z13.discriminatedUnion("type", [
      *  "call me back after you merge" in their own words decides for itself what to do,
      *  and now knows exactly which call to make. Absent when either half is missing —
      *  a sentence with a hole in it is worse than no sentence. */
-    note: z13.string().optional(),
+    note: z14.string().optional(),
     /** The call record rendered for THIS agent (`docs/brain/voice/record-design.md`): the words the
      *  shaped answer was mapped from, filtered to its own claims. There is no second list
      *  of labels beside it — the acts went 2026-09-04 and `intents` went with them (owner,
      *  2026-09-04): the agent reads the sentence and decides. */
-    transcript: z13.string().optional(),
+    transcript: z14.string().optional(),
     /** Coverage report (#396), when the ask declared `points`: which of them this
      *  answer addressed. Missing points = re-ask or proceed knowingly partial. */
-    covered: z13.array(z13.string()).optional(),
+    covered: z14.array(z14.string()).optional(),
     /** Ride-alongs (RideAlongSchema) — pending work for you, attached to the moment you
      *  became free. Only `reply` and `idle` carry it: those are the two outcomes that
      *  END a wait. `remind`, `superseded` and `turn` are mid-flight, and handing an
      *  agent a side-quest while it is still holding the line is how the main thing gets
      *  dropped. Absent/empty = nothing owed. */
-    also: z13.array(RideAlongSchema).optional()
+    also: z14.array(RideAlongSchema).optional()
   }),
-  z13.object({
-    type: z13.literal("remind"),
-    parentId: z13.string(),
-    notificationId: z13.string(),
-    remindAt: z13.string().datetime({ offset: true }),
+  z14.object({
+    type: z14.literal("remind"),
+    parentId: z14.string(),
+    notificationId: z14.string(),
+    remindAt: z14.string().datetime({ offset: true }),
     /** Seconds until remindAt, server-computed — pass straight to ScheduleWakeup. */
-    remindInSeconds: z13.number()
+    remindInSeconds: z14.number()
   }),
   /** The awaited ask was REPLACED by a newer notification on its thread (e.g. a
    *  post-feedback revision, #633) — the user will never answer this id. Stop
    *  awaiting it; the live ask is the thread's newest turn (await that one, or
    *  re-orient via contact({})). */
-  z13.object({
-    type: z13.literal("superseded"),
-    parentId: z13.string(),
-    notificationId: z13.string()
+  z14.object({
+    type: z14.literal("superseded"),
+    parentId: z14.string(),
+    notificationId: z14.string()
   }),
   /** A LIVE call's turn, streamed as it lands (#783). PROVISIONAL: the user can still
    *  revise any of these until the final reply arrives — partial = intelligence,
    *  settled = authorization. Use it to PREPARE (fetch, draft, warm), never to act
    *  irreversibly. If `acts` carries a question aimed at you and you know the answer,
    *  contact on the same thread right away — the caller hears it on the same call. */
-  z13.object({
-    type: z13.literal("partial"),
-    notificationId: z13.string(),
-    inFlight: z13.literal(true),
-    turn: z13.object({
-      idx: z13.number(),
-      prompt: z13.string(),
-      reply: z13.string(),
-      acts: z13.array(IntentSchema).nullable().optional()
+  z14.object({
+    type: z14.literal("partial"),
+    notificationId: z14.string(),
+    inFlight: z14.literal(true),
+    turn: z14.object({
+      idx: z14.number(),
+      prompt: z14.string(),
+      reply: z14.string(),
+      acts: z14.array(IntentSchema).nullable().optional()
     })
   }),
-  z13.object({
-    type: z13.literal("idle"),
-    also: z13.array(RideAlongSchema).optional(),
+  z14.object({
+    type: z14.literal("idle"),
+    also: z14.array(RideAlongSchema).optional(),
     /** Is a call live for this agent's user right now? The SDK polls the partial stream
      *  (#783) between idle ticks ONLY while this is not `false` — a partial can only exist
      *  during a live call, and polling for one on a banner/message was a wasted HTTP call +
      *  3 queries on every idle tick of every waiting agent (~80% of all traffic at scale).
      *  Absent = an older API → the SDK keeps polling, exactly as before. */
-    inFlight: z13.boolean().optional()
+    inFlight: z14.boolean().optional()
   })
 ]);
-var VoiceKeySchema = z13.enum(["rachel", "george", "jessica", "brian", "lily"]);
-var AgendaTurnSchema = z13.object({
+var VoiceKeySchema = z14.enum(["rachel", "george", "jessica", "brian", "lily"]);
+var AgendaTurnSchema = z14.object({
   /** THE TURN'S IDENTITY (the first-sentence stream, 2026-09-09): the brain call that wrote
    *  it and its place in that reply — `<brainCallId>:<index>`, with `:p` on the first
    *  sentence a re-plan publishes ahead of the rest. A turn is spoken once, by this id: the
    *  completion of a streamed re-plan carries the published sentence again, and the walk
    *  drops what it already said by identity, never by the API's guess of what was polled.
    *  Absent on plans nothing streams (a ring plan, a floor). */
-  id: z13.string().optional(),
+  id: z14.string().optional(),
   /** Twin coverage (#1089): sibling claim ids this asking turn's answer ALSO settles —
    *  the planner declares duplicates instead of asking them twice. */
-  coveredIds: z13.array(z13.string()).optional(),
+  coveredIds: z14.array(z14.string()).optional(),
   /** The spoken sentences of the turn, in order. No count: how long a turn is is the brain's call
    *  (owner, 2026-09-25), and a count here refused whole plans. */
-  info: z13.array(z13.string().min(1)).default([]),
+  info: z14.array(z14.string().min(1)).default([]),
   /** THE ID OF EACH SENTENCE, when the plan names them (turns.md): a talker message's sentences are its
    *  `info`, one piece each, and `units[i]` is the id the bot's line receipt carries for `info[i]`
    *  (`<cycle>:<key>:<n>`), so what was heard is known per sentence. Absent on every other turn. */
-  units: z13.array(z13.string().min(1)).optional(),
-  question: z13.string().min(1).nullable(),
+  units: z14.array(z14.string().min(1)).optional(),
+  question: z14.string().min(1).nullable(),
   /** True on the one turn carrying the agent's own declared question. */
-  asks: z13.boolean().optional(),
+  asks: z14.boolean().optional(),
   /** The claim this turn belongs to (#781) — the RETURN identity: answers route by it.
    *  Absent on a single-claim plan (the session's own claim) and on shared context turns,
    *  which route nothing. */
-  claimId: z13.string().optional(),
+  claimId: z14.string().optional(),
   /** The claim's voice key (#462) — the OUTBOUND identity, audible who-is-asking. */
-  voice: z13.string().optional(),
+  voice: z14.string().optional(),
   /** The claim's AGENT NAME (#838) — the spoken identity. A voice alone doesn't say
    *  whose request this is: an item that folded in from another agent arrived as a bare
    *  non-sequitur ("First real production sign-in is yours to make whenever you want.")
    *  and the owner answered "What?". The bot names the agent before its first turn. */
-  agent: z13.string().optional(),
+  agent: z14.string().optional(),
   /** The claim's agent by ID — the pairing's connection id (`notifications.token_id`), the
    *  same id a face is minted from. A name is not an identity: two pairings may be called
    *  "Claude", and a name cannot be joined on. The record's entries carry it (`agent_id`)
    *  so "who said that" survives the call, and it rides PER TURN because a coalesced call
    *  speaks for several agents — the turn is the only place that knows which. */
-  agentId: z13.string().optional(),
+  agentId: z14.string().optional(),
   /** ITS QUESTION IS ON THE TABLE (#3292): a talker line about a question the person owes. The walk tells
    *  the phone (`asking`, so the agenda goes gold and its options show) without treating the line as a
    *  question turn — the talker owns re-asking and the waits, so `question` stays null. */
-  onTable: z13.boolean().optional(),
+  onTable: z14.boolean().optional(),
   /** THE CARRIED ITEM THIS LINE TELLS (Goal 2a1e50eb, owner 2026-10-14: "Build the whole design") — a
    *  request the Call carries that asks NOTHING: a report, a progress note, another agent's waiting news.
    *  Its request Entry id, so the phone can make that agenda row the current one while it is read
@@ -1222,14 +1309,14 @@ var AgendaTurnSchema = z13.object({
    *  record tags the line with it (`workId`) — and an answer must never route to an Entry that asks
    *  nothing. Absent on a turn speaking for a question (`claimId` says that) and on one speaking for no
    *  carried item at all (an opener, a sign-off, Paigy's own line). */
-  telling: z13.string().optional(),
+  telling: z14.string().optional(),
   /** PAIGY'S OWN LINE (owner, 2026-10-15; calls ed2cbdce, ad50ad25): a turn that speaks for no item -- an
    *  opener, a sign-off, a reply about nothing an agent asked or told -- is recorded as Paigy's, not as the
    *  agent that rang. Absent on every turn that speaks for an item (`agentId` names whose), and on plans
    *  from before it: a bot that does not know it records the ringing agent, as before. */
-  speaksFor: z13.literal("paigy").optional(),
+  speaksFor: z14.literal("paigy").optional(),
   select: SelectShapeSchema.optional(),
-  options: z13.array(OptionSchema.omit({ id: true })).optional(),
+  options: z14.array(OptionSchema.omit({ id: true })).optional(),
   /* `pace` STOOD HERE (#826). A turn could carry seconds and the model chose them. The walk
      paces itself now — a short beat between the sentences of a turn, the longer one at its end
      (owner, 2026-09-30: "remove the bot deciding pace") — and it does that where the words are
@@ -1238,58 +1325,58 @@ var AgendaTurnSchema = z13.object({
   /** Whether the walk WAITS for an answer before moving on. Absent = derived as today
    *  (a question blocks, context flows). blocking:false on a question = ask and move
    *  on, the claim stays pending; blocking:true on context = hold for a reply. */
-  blocking: z13.boolean().optional(),
+  blocking: z14.boolean().optional(),
   /** SPOKEN ONLY IF THEY SAY NOTHING (owner, 2026-10-01, call 812de935: "you're gonna re-ask, but it
    *  shouldn't be the same words … more like, hey, are you still there, or are you able to answer, or
    *  would you need more information"). The walk holds this turn out of its queue; at the queue's end it
    *  listens for the last word, and only if that listen is silent is this turn said and asked. If they
    *  speak, it is dropped and their words are taken like any reply. */
-  ifSilent: z13.boolean().optional()
+  ifSilent: z14.boolean().optional()
 });
 var CLAIM_STALE_MS = 30 * 6e4;
-var CallQuestionSchema = z13.object({
-  id: z13.string(),
-  entryId: z13.string(),
-  title: z13.string(),
-  state: z13.enum(["open", "answered"]),
-  answer: z13.string().nullable(),
+var CallQuestionSchema = z14.object({
+  id: z14.string(),
+  entryId: z14.string(),
+  title: z14.string(),
+  state: z14.enum(["open", "answered"]),
+  answer: z14.string().nullable(),
   /** WHO ASKED IT (owner, 2026-09-23, Goal a345e906: each agenda row wears its agent's face) — the
    *  request Entry's author, as the same three facts the item's own `tokenId`/`name`/`voice`
    *  carry for the call's one agent, so the phone draws it with the same seed. Absent when the
    *  author is not an agent this account holds (unpaired since, or a person). */
-  agent: z13.object({ tokenId: z13.string(), name: z13.string(), voice: VoiceKeySchema.optional() }).optional(),
+  agent: z14.object({ tokenId: z14.string(), name: z14.string(), voice: VoiceKeySchema.optional() }).optional(),
   /** ITS OPTIONS, WHEN THERE IS SOMETHING TO SEE (owner, 2026-09-25: "Yes, add it"): the options
    *  its need offers, exactly as its own card carries them, present only when one of them has a
    *  preview (`html` or `image`). The call screen opens them from the agenda row, so a preview is
    *  never re-sent as a second card to be seen mid-call. Words-only options are `choices`. */
-  options: z13.array(OptionSchema).optional(),
+  options: z14.array(OptionSchema).optional(),
   /** ITS OPTIONS WHEN THEY ARE WORDS (owner, 2026-10-02, on a call: "whenever there are pre-made
    *  options … now for a checklist, we should also show them on the screen"; design C, 2026-10-08):
    *  numbered chips above the call's controls, so a checklist is never only something read aloud. A
    *  field of its own so `options` keeps meaning "previews": a phone on an older bundle opens the
    *  preview layout for any `options`, and must see exactly what it did. */
-  choices: z13.array(OptionSchema).optional(),
+  choices: z14.array(OptionSchema).optional(),
   /** HOW MANY MAY BE PICKED, beside its `choices`: one, any (`many`) or an order (`rank`). */
-  select: z13.enum(["one", "many", "rank"]).optional()
+  select: z14.enum(["one", "many", "rank"]).optional()
 });
-var InboxItemSchema = z13.object({
-  id: z13.string(),
-  tokenId: z13.string().optional(),
+var InboxItemSchema = z14.object({
+  id: z14.string(),
+  tokenId: z14.string().optional(),
   status: NotifyStatusSchema,
   context: ContextSchema,
-  options: z13.array(OptionSchema).optional(),
+  options: z14.array(OptionSchema).optional(),
   /** The ask's declared coverage points (#396), when the agent sent them. */
-  points: z13.array(z13.string()).optional(),
+  points: z14.array(z14.string()).optional(),
   /** Does this claim want an ANSWER, or is it telling you something? Written per row from
    *  `requestAsks` — the agent's own declaration, not a guess. `false` is what earns a card
    *  its acknowledge affordance: without it a status update offers a text box and a dismiss,
    *  and neither of those is "got it" (owner, 2026-08-10). */
-  asks: z13.boolean().optional(),
+  asks: z14.boolean().optional(),
   /** When a live process last pulsed for this row's agent — the liveness input for
    *  "working requires a pulse" (#928): the list said "Working…" from agent_state alone
    *  while the party called the same dead claim stalled. Absent = no token/no data,
    *  which must never CLAIM stalled. */
-  lastSeenAt: z13.string().optional(),
+  lastSeenAt: z14.string().optional(),
   /** WHEN THE AGENT LAST SAID ANYTHING ABOUT THIS CLAIM — the newest `agent_state` row in
    *  the `notification_events` ledger (trigger-written since 20260621010000, so every row a
    *  user can see has one). The age input for `CLAIM_STALE_MS`, and it has to be this rather
@@ -1299,7 +1386,7 @@ var InboxItemSchema = z13.object({
    *  work. Reading the row's birth as the claim's age brands that "No update in 8h" the
    *  instant the agent picks it up (#997). Absent = pre-trigger row; fall back to
    *  `createdAt`. */
-  agentStateAt: z13.string().datetime().optional(),
+  agentStateAt: z14.string().datetime().optional(),
   /** THE QUESTIONS A CALL CARRIES — the call screen's agenda spine (docs/clients/app/walk/design.md §11, owner
    *  2026-09-22). One per DecisionNeed on the Call, in the Call's order, answered or open (a
    *  superseded or cancelled need is no longer a question anyone is asked). Present only on a
@@ -1313,35 +1400,39 @@ var InboxItemSchema = z13.object({
    *  `turn` topic (`asking`, `settled`), because the bot never sees a DecisionNeed id. `title` is
    *  the card's own concise heading; `answer` the accepted answer in words, null while open. It
    *  REPLACED `agenda` (turns), which nothing ever filled. */
-  questions: z13.array(CallQuestionSchema).optional(),
+  questions: z14.array(CallQuestionSchema).optional(),
   /** The same call carrying order, including request Entries with no DecisionNeed as reports.
    *  Additive: questions stays decision-only for every other surface and older bundles. */
-  callItems: z13.array(CallQuestionSchema.extend({ state: z13.enum(["open", "answered", "report"]), acknowledged: z13.boolean().optional() })).optional(),
-  visuals: z13.array(VisualSchema).optional(),
+  callItems: z14.array(CallQuestionSchema.extend({ state: z14.enum(["open", "answered", "report"]), acknowledged: z14.boolean().optional() })).optional(),
+  visuals: z14.array(VisualSchema).optional(),
   /** The connected agent's name (the single pairing name — user-typed, or the
    *  agent's suggestion, or a default silly name). */
-  name: z13.string(),
+  name: z14.string(),
   /** The pairing's assigned voice (#462); absent = the default voice. */
   voice: VoiceKeySchema.optional(),
-  repo: z13.string().optional(),
-  branch: z13.string().optional(),
-  createdAt: z13.string().datetime(),
-  snoozedUntil: z13.string().datetime().optional(),
+  repo: z14.string().optional(),
+  branch: z14.string().optional(),
+  createdAt: z14.string().datetime(),
+  snoozedUntil: z14.string().datetime().optional(),
+  /** A past card the person dismissed that Restore can take back (#2699): its newest close is their
+   *  dismissal, with what it withdrew recorded (`private.restore_delivery`). Absent on everything else —
+   *  a put-off, a card an ended Goal or a settled question closed, and every open card. */
+  restorable: z14.boolean().optional(),
   agentState: AgentStateSchema.default("idle"),
   /** Whose action the item is waiting on: "you" = an agent asked you (the default,
    *  every agent→user notification); "agent" = you sent a request and it's awaiting the
    *  agent (held in the inbox until the agent replies on the thread). */
-  turn: z13.enum(["you", "agent"]).default("you"),
+  turn: z14.enum(["you", "agent"]).default("you"),
   /** Hard error reason on an awaiting request (turn="agent") — the wake failed to reach
    *  the agent (provider-agnostic; set server-side). Absent = no hard error. Drives the inbox
    *  error badge + Retry. */
-  error: z13.string().optional(),
-  clarifies: z13.string().optional(),
+  error: z14.string().optional(),
+  clarifies: z14.string().optional(),
   /** THIS CARD'S QUESTION IS ON A LIVE CALL (owner, 2026-09-24: "Mark it while the call is
    *  live"). Present only while an open Call Delivery carries the card's request Entry — read
    *  off the same open list the card came from, so it clears when the Call does. A card is the
    *  backup for a call not taken; while the call has it, the call is where it is answered. */
-  onCall: z13.literal(true).optional(),
+  onCall: z14.literal(true).optional(),
   /** THE RING, ON THE ITEM (docs/clients/app/walk/design.md §12 §17, #2251): the last ring on this card was
    *  declined, and what the ladder will do next — read off the cron's own row, never computed
    *  on the phone. Present only while a `declined` receipt stands on the card's last Call.
@@ -1351,31 +1442,31 @@ var InboxItemSchema = z13.object({
    *  It replaced `gaveUp` (deleted 2026-09-22): "the ladder spent" was a boolean the projection
    *  never set, and it is `nextRingAt === null` here — the party's *Missed you* (`party/dress.ts`)
    *  and the roster's `unreached` read `declinedAt`, and stand while it does. */
-  ring: z13.object({
-    declinedAt: z13.string().datetime(),
-    anchorAt: z13.string().datetime(),
-    nextRingAt: z13.string().datetime().nullable(),
-    step: z13.number().int()
+  ring: z14.object({
+    declinedAt: z14.string().datetime(),
+    anchorAt: z14.string().datetime(),
+    nextRingAt: z14.string().datetime().nullable(),
+    step: z14.number().int()
   }).optional(),
   /** Why this arrived the way it did, read back off the delivery receipt (`notify/why.ts`).
    *  Absent for anything never delivered through a push, and for older rows written before
    *  the reason was recorded. Deliberately a debug affordance, shown small (owner,
    *  2026-08-07) — its real job is to give "this didn't need a call" something to be
    *  feedback ABOUT. */
-  why: z13.object({
+  why: z14.object({
     asked: NotifyLevelSchema,
     got: NotifyLevelSchema,
-    because: z13.enum(["unresponsive", "dismissed", "not_permitted", "silent", "coalesced", "agent_capped", "unplanned", "learned_raise"]).optional(),
-    line: z13.string()
+    because: z14.enum(["unresponsive", "dismissed", "not_permitted", "silent", "coalesced", "agent_capped", "unplanned", "learned_raise"]).optional(),
+    line: z14.string()
   }).optional(),
-  select: z13.enum(["one", "many", "rank", "confirm", "text"]).default("one"),
-  confirmStyle: z13.enum(["yesno", "approve"]).default("yesno").describe(
+  select: z14.enum(["one", "many", "rank", "confirm", "text"]).default("one"),
+  confirmStyle: z14.enum(["yesno", "approve"]).default("yesno").describe(
     "Labels for a select:'confirm' paige \u2014 'yesno' (Yes/No) or 'approve' (Approve/Deny). Ignored unless select is 'confirm'."
   ),
   /** Real downstream work is stuck behind this one — set by the agent, independent of
    *  urgency (see the main README's "premier use case" + docs/history/notify-states.md). Drives the
    *  inbox's blocking badge and the extra confirm step before dismissing it. */
-  blocking: z13.boolean().default(false),
+  blocking: z14.boolean().default(false),
   /** WHAT FREED THIS QUESTION (owner, 2026-10-16, Goal 2383ef14: "we don't really show the question
    *  at the top until the question it could reply to with is resolved. And then we show it again
    *  with the context of the other thing that was resolved"; asked how it should carry that, "The
@@ -1383,25 +1474,25 @@ var InboxItemSchema = z13.object({
    *  (`private.ask_back`) and drops out of the front while it waits; when it comes back, the card
    *  draws their question and the answer it got ABOVE the ask, as one card. Absent on a question
    *  nothing ever held back. */
-  freedBy: z13.object({
-    questionId: z13.string(),
-    asked: z13.string(),
-    answer: z13.string(),
-    answeredAt: z13.string(),
+  freedBy: z14.object({
+    questionId: z14.string(),
+    asked: z14.string(),
+    answer: z14.string(),
+    answeredAt: z14.string(),
     /** The answer in ten words, when the talker wrote one beside it. */
-    summary: z13.string().optional()
+    summary: z14.string().optional()
   }).optional(),
   /** WHAT THIS CARD ANSWERS (owner, 2026-10-27, Goal 81b6a9d6: "this reads very clearly as an answer to
    *  something I asked, but I don't see the question… To show the question near the top of the card").
    *  An agent's card that is its answer to one of the person's own questions carries that question's
    *  words, drawn above it as "You asked" (`public.questions_answered_by`, read beside the card read).
    *  Absent on every card that answers nothing of theirs, and on a held question, which has `freedBy`. */
-  answering: z13.object({ asked: z13.string() }).optional(),
+  answering: z14.object({ asked: z14.string() }).optional(),
   /** WHEN ITS AGENT LAST BUMPED THIS QUESTION (#2855, `contact`'s `bumpQuestionIds`): the agent asked
    *  the person to notice it again, and For you puts it first, the latest bump on top
    *  (`question/order.ts`). Absent on a question never bumped, and on every card from an API older
    *  than the bump. */
-  bumpedAt: z13.string().optional(),
+  bumpedAt: z14.string().optional(),
   /** The user's locked-in answer; present only for replied items (GET /api/replied/:id). */
   answer: UserAnswerSchema.optional(),
   /** THE TARGET FACTS A CARD RENDERS (#1796 point 5, 2026-09-11): the Delivery it is a view of,
@@ -1409,17 +1500,17 @@ var InboxItemSchema = z13.object({
    *  for a request that asks nothing), whether its content is sealed, and that Goal's state. The
    *  answer writer (`POST /api/entries`) and the disposition (`close_delivery`) take their ids from
    *  here. The server projects it (`apps/api/src/inbox/project.ts`); a client never builds it. */
-  communication: z13.object({
-    deliveryId: z13.string(),
-    kind: z13.enum(["notification", "call"]),
-    entryId: z13.string(),
-    goalIds: z13.array(z13.string()),
-    decisionNeedId: z13.string().optional(),
-    sealed: z13.boolean(),
-    goalState: z13.string().optional(),
+  communication: z14.object({
+    deliveryId: z14.string(),
+    kind: z14.enum(["notification", "call"]),
+    entryId: z14.string(),
+    goalIds: z14.array(z14.string()),
+    decisionNeedId: z14.string().optional(),
+    sealed: z14.boolean(),
+    goalState: z14.string().optional(),
     /** THAT GOAL'S NAME (#2416) — what Activity's row is headed by, since a row there is one Goal
      *  and the cards it holds sit behind it. Stamped by the same read as `goalState`. */
-    goalTitle: z13.string().optional()
+    goalTitle: z14.string().optional()
   }).optional(),
   /** WHAT THIS CARD IS, IN TWELVE CHARACTERS (#3019) — the hash of every other field on it, stamped
    *  by the one read that serves the open list (`apps/api/src/inbox/project.ts` `inboxFor`). It is
@@ -1431,27 +1522,57 @@ var InboxItemSchema = z13.object({
    *  own moves (its Goal's state and name, whether a ring is
    *  live). Optional, so a fixture, the demo and the archive lens need not spell one, and a card
    *  with no rev is simply always re-sent. */
-  rev: z13.string().optional()
+  rev: z14.string().optional()
 });
 var APNS_TOKEN_RE = /^[0-9a-fA-F]{64}$/;
-var PushTokenSchema = z13.object({
-  voipToken: z13.string().min(1).optional(),
-  alertToken: z13.string().min(1).optional(),
-  fcmToken: z13.string().min(1).optional(),
-  platform: z13.enum(["ios", "android"])
+var PushDeviceIdSchema = z14.string().regex(/^[A-Za-z0-9._:-]{1,200}$/, "not an install id");
+var PushDeviceSchema = z14.object({
+  deviceId: z14.string(),
+  platform: z14.enum(["ios", "android"]),
+  label: z14.string().nullable(),
+  model: z14.string().nullable(),
+  /** Calls and banners go here. */
+  rings: z14.boolean(),
+  /** It has a token a call can ring. */
+  ringable: z14.boolean(),
+  chosenAt: z14.string().nullable(),
+  updatedAt: z14.string()
+});
+var PushTokenDeviceSchema = z14.object({
+  deviceId: z14.string(),
+  rings: z14.boolean(),
+  ask: z14.boolean(),
+  ringsOn: z14.object({ deviceId: z14.string().nullable(), label: z14.string().nullable() }).nullable()
+});
+var PushRingsSchema = z14.object({ rings: z14.boolean() });
+var PushTokenSchema = z14.object({
+  voipToken: z14.string().min(1).optional(),
+  alertToken: z14.string().min(1).optional(),
+  fcmToken: z14.string().min(1).optional(),
+  platform: z14.enum(["ios", "android"]),
+  /** WHICH INSTALL POSTED IT (owner, 2026-11-05: "Ask on each new device"). Absent from every
+   *  build before it, and then the post is exactly what it always was: the one row per account.
+   *  With it, the install has its own row and calls ring only where the person chose. */
+  deviceId: PushDeviceIdSchema.optional(),
+  /** The install's name as the person knows it ("iPhone", "iPad"), and its model. */
+  label: z14.string().trim().min(1).max(80).optional(),
+  model: z14.string().trim().min(1).max(80).optional(),
+  /** The person's choice, when the post carries it; absent, the install keeps what it has (and
+   *  its first ring token decides once: it rings unless another install already does). */
+  rings: z14.boolean().optional()
 }).superRefine((v, ctx) => {
   if (v.platform !== "ios") return;
   for (const field of ["voipToken", "alertToken"]) {
     const token = v[field];
     if (token === void 0 || APNS_TOKEN_RE.test(token)) continue;
     ctx.addIssue({
-      code: z13.ZodIssueCode.custom,
+      code: z14.ZodIssueCode.custom,
       path: [field],
       message: `not an APNs device token (want 64 hex chars, got ${token.length})`
     });
   }
 });
-var MissedCallSchema = z13.enum([
+var MissedCallSchema = z14.enum([
   "retry_10m",
   "retry_30m",
   "retry_60m",
@@ -1464,31 +1585,31 @@ var MissedCallSchema = z13.enum([
 var clock = (h) => h === 0 ? "midnight" : h === 12 ? "noon" : h < 12 ? `${h} am` : `${h - 12} pm`;
 var QUIET = ` Nothing rings from ${clock(NIGHT.from)} to ${clock(NIGHT.to)} your time; the count waits for morning.`;
 var THEIR_QUIET = ` Nothing rings from ${clock(NIGHT.from)} to ${clock(NIGHT.to)} their time; the count waits for morning.`;
-var BrokerTuningSchema = z13.object({
+var BrokerTuningSchema = z14.object({
   /** 'none' = skip the spoken ack after a mapped answer (power users find it slow). */
-  ackVerbosity: z13.enum(["normal", "none"]).optional(),
+  ackVerbosity: z14.enum(["normal", "none"]).optional(),
   /** How readily the mapper asks its one clarification: 'low' = only when truly
    *  uninterpretable, 'high' = whenever not fully certain. */
-  clarifyEagerness: z13.enum(["low", "normal", "high"]).optional(),
+  clarifyEagerness: z14.enum(["low", "normal", "high"]).optional(),
   /** The user's own shorthand: when they say `say`, they mean `mean`. */
-  phrasebook: z13.array(z13.object({ say: z13.string().min(1).max(60), mean: z13.string().min(1).max(120) })).max(24).optional(),
+  phrasebook: z14.array(z14.object({ say: z14.string().min(1).max(60), mean: z14.string().min(1).max(120) })).max(24).optional(),
   /** The language calls are PLANNED in, when the account has chosen one (#1272). Absent —
    *  which is every account today — means the agent's own words decide, per ask: a call
    *  about an English ask opens in English. This is the only thing that overrides that,
    *  and a live caller who switches language mid-call still outranks it (broker/lang.ts).
    *  Set per user (no UI yet), like `voiceTuning`. */
-  language: z13.enum(["en", "es"]).optional()
+  language: z14.enum(["en", "es"]).optional()
 });
-var UserSettingsSchema = z13.object({
-  permissions: z13.object({
-    call: z13.boolean(),
-    banner: z13.boolean(),
-    push: z13.boolean()
+var UserSettingsSchema = z14.object({
+  permissions: z14.object({
+    call: z14.boolean(),
+    banner: z14.boolean(),
+    push: z14.boolean()
   }),
   /** LockedIn / Default / DateNight on screen; the stored words are unchanged on purpose —
    *  they are an enum on a live column across every account, and the rename is a rename of
    *  what people read (owner, 2026-09-30). */
-  sessionMode: z13.enum(["default", "all_calls", "silent"]),
+  sessionMode: z14.enum(["default", "all_calls", "silent"]),
   /** `silentPush` lived here until #2813 and is now GONE, field and column both. It was kept as an
    *  optional long after DateNight stopped reading it, on the theory that a phone on an older
    *  bundle PATCHing the whole settings object would be REFUSED for sending a key we had stopped
@@ -1497,7 +1618,7 @@ var UserSettingsSchema = z13.object({
    *  an old bundle's `silentPush` is accepted and ignored. Worth remembering before keeping the
    *  next dead field for the same reason. */
   /** Opt-in (default false) to using your content to improve Paigy and train models. */
-  improveConsent: z13.boolean(),
+  improveConsent: z14.boolean(),
   missedCall: MissedCallSchema.default("backoff_standard"),
   /** Where voice audio is processed. 'hosted' (default) = Paigy's voice services
    *  (ElevenLabs TTS, faster-whisper STT, the call bot); 'on_device' = the phone
@@ -1505,14 +1626,14 @@ var UserSettingsSchema = z13.object({
    *  Optional, NOT defaulted: a stale client PATCHing the full settings object
    *  must not silently reset this privacy choice. Absent = leave unchanged on
    *  write, 'hosted' on read (see store.ts). */
-  voiceMode: z13.enum(["hosted", "on_device"]).optional(),
+  voiceMode: z14.enum(["hosted", "on_device"]).optional(),
   /** `talk` lived here until #2373: it gated auto-play, which went in #2374. An older bundle's
-   *  `talk` is stripped like `silentPush` was; the `user_settings.talk` column outlives it until
-   *  its own drop migration (#2373, step two). */
+   *  `talk` is stripped like `silentPush` was, and the `user_settings.talk` column went in
+   *  20261028142255. */
   /** CALL DIAGNOSTICS (owner, 2026-10-01): the call report carries each listen and the bot's own
    *  load timings. SERVER-SET, no UI — on for every account that existed on 2026-10-01, off for
    *  newer ones (migration 20261001132859). Read-only here: the settings PATCH never writes it. */
-  callDiagnostics: z13.boolean().optional(),
+  callDiagnostics: z14.boolean().optional(),
   /** Per-user ring budget (#603): calls per rolling day before further calls
    *  degrade to banner. Absent = the global default (25). A number, never a
    *  bypass — every account keeps a ceiling. No UI; set per user for testing. */
@@ -1520,7 +1641,7 @@ var UserSettingsSchema = z13.object({
    *  payload['tuning'] (e.g. { silence_s: 3.5 } — a longer pause window for a
    *  slower speaker). No API-side semantics; the bot resolves each key with its
    *  own defaults. Set per user (no UI yet); absent = bot defaults. */
-  voiceTuning: z13.record(z13.string(), z13.union([z13.number(), z13.string()])).optional(),
+  voiceTuning: z14.record(z14.string(), z14.union([z14.number(), z14.string()])).optional(),
   /** The user's IANA timezone (e.g. "America/Bogota"), recorded by the app — it is the
    *  only party that knows it. REMINDERS are why it exists: "remind me at ten" becomes
    *  an absolute `due_at` only if we know whose ten. Optional and never defaulted, for
@@ -1529,71 +1650,75 @@ var UserSettingsSchema = z13.object({
    *  that failure reads as the reminder rail being unreliable rather than as a missing
    *  setting. Absent = a spoken time can't be landed, so the reminder rides the next
    *  call — honest about what we know. */
-  timezone: z13.string().min(1).max(64).optional(),
+  timezone: z14.string().min(1).max(64).optional(),
   /** Rung-2 broker tuning (#381). Optional and NOT defaulted, same stale-client
    *  clobber guard as voiceMode: absent = leave unchanged on write. */
   broker: BrokerTuningSchema.optional()
 });
-var GoalLatestSchema = z13.object({
-  from: z13.enum(["person", "agent"]),
-  said: z13.string(),
-  at: z13.string(),
-  entryId: z13.string(),
+var GoalLatestSchema = z14.object({
+  from: z14.enum(["person", "agent"]),
+  said: z14.string(),
+  at: z14.string(),
+  entryId: z14.string(),
   /** ITS SUMMARY IN AT MOST TEN WORDS (owner, 2026-10-08: "the same switch to the more summarized sources
    *  at the work tab top level screen too"): what the talker or filer wrote on the Entry; absent when
    *  none was. The Work row reads it before `said`. */
-  summary: z13.string().optional(),
+  summary: z14.string().optional(),
   /** AN AGENT'S UPDATE OR QUESTION AS ITS UNITS' TITLES (owner, 2026-10-08: "send the 50 chars from the title
    *  of each unit"): each at most 50 characters, in order; absent when it has none. */
-  titles: z13.array(z13.string()).optional(),
+  titles: z14.array(z14.string()).optional(),
   /** A PROGRESS NOTE'S ONE POINT ("An agent's progress note -> maybe let's cap it at 200 chars"): the
    *  update's own line, before its units; the row caps it at 200. Absent on anything else. */
-  line: z13.string().optional()
+  line: z14.string().optional(),
+  /** THE PERSON'S ENTRIES IT CARRIED TO AN AGENT, IN THEIR WORDS (owner, 2026-11-05: "paraphrasing shouldn't change
+   *  attribution in the UI"): on a message Paigy sent an agent, the person's Entries its Delivery relays whole. A row
+   *  draws it as theirs. Absent on anything else, and from a door older than it. */
+  carries: z14.array(z14.string()).optional()
 });
-var HistoryWorkSchema = z13.object({
-  id: z13.string(),
-  title: z13.string(),
-  state: z13.enum(["done", "cancelled"]),
+var HistoryWorkSchema = z14.object({
+  id: z14.string(),
+  title: z14.string(),
+  state: z14.enum(["done", "cancelled"]),
   /** Who held it (`agent:<tokenId>` or `human:<userId>`). */
-  assignee: z13.string(),
+  assignee: z14.string(),
   latest: GoalLatestSchema.nullable().optional()
 });
-var HistoryEntrySchema = z13.union([
-  z13.object({ at: z13.string(), card: InboxItemSchema }),
-  z13.object({ at: z13.string(), work: HistoryWorkSchema })
+var HistoryEntrySchema = z14.union([
+  z14.object({ at: z14.string(), card: InboxItemSchema }),
+  z14.object({ at: z14.string(), work: HistoryWorkSchema })
 ]);
-var HistoryPageSchema = z13.object({
-  entries: z13.array(HistoryEntrySchema),
-  next: z13.string().nullable()
+var HistoryPageSchema = z14.object({
+  entries: z14.array(HistoryEntrySchema),
+  next: z14.string().nullable()
 });
 var ACTIVITY_LINES = 2;
 var ACTIVITY_LINE_MAX = 80;
-var AgentActivitySchema = z13.object({
+var AgentActivitySchema = z14.object({
   /** Oldest first, so the newest line is last — the one that replaces in place. */
-  lines: z13.array(z13.string().max(ACTIVITY_LINE_MAX)).max(ACTIVITY_LINES),
+  lines: z14.array(z14.string().max(ACTIVITY_LINE_MAX)).max(ACTIVITY_LINES),
   /** When the harness observed this tail. Its own timestamp, not the heartbeat's: a beat
    *  that carries an UNCHANGED tail must not make a stalled agent look like it just moved. */
-  at: z13.string().datetime()
+  at: z14.string().datetime()
 });
-var ConnectionSummarySchema = z13.object({
+var ConnectionSummarySchema = z14.object({
   /** The connection = the agent's token id (used to address a request). */
-  id: z13.string(),
+  id: z14.string(),
   /** The credential kind: "device" = a paired machine (mint-only — it hosts and mints, it
    *  never talks); "agent" = an identity that sends. The roster and devices surfaces split
    *  on this. Optional/absent reads as "agent" (a row predating the kind column). See
    *  docs/server/tokens/devices-vs-agents-design.md. */
-  kind: z13.enum(["device", "agent"]).optional(),
+  kind: z14.enum(["device", "agent"]).optional(),
   /** For an agent, the token id of the DEVICE that minted it — so agents group under their
    *  machine, and revoking a device cascades to them. Null on devices, and on unlinked
    *  agents (provider-managed, hatched by another agent, or minted before the link existed);
    *  a phone-launched agent is linked to the computer it runs on since #1602. */
-  mintedByDevice: z13.string().nullable().optional(),
-  device: z13.string().nullable(),
+  mintedByDevice: z14.string().nullable().optional(),
+  device: z14.string().nullable(),
   /** The agent's display name (the single pairing name). */
-  name: z13.string(),
+  name: z14.string(),
   /** For a managed connection, the provider key (e.g. "cma") that agentOrigin maps to a
    *  label; null for a local connection. Sourced from the token's provider, not the name. */
-  provider: z13.string().nullable(),
+  provider: z14.string().nullable(),
   /** The pairing's assigned voice (#462); null = the default voice. */
   voice: VoiceKeySchema.nullable(),
   /** The LOUDEST this agent may ever reach you — a ceiling on `NOTIFY_LADDER`, set by the
@@ -1604,34 +1729,40 @@ var ConnectionSummarySchema = z13.object({
    *  every surface at once and outranks even `sessionMode: all_calls` — a mode the user
    *  set once must not overrule a rule they set about one agent. */
   reach: NotifyLevelSchema.nullable().optional(),
-  createdAt: z13.string().datetime(),
+  createdAt: z14.string().datetime(),
   /** Most recent notification on this connection, either direction. Null = no contact yet.
    *  Drives the agents-page recency grouping (Today / This week / …). */
-  lastContactAt: z13.string().datetime().nullable(),
+  lastContactAt: z14.string().datetime().nullable(),
   /** Last presence heartbeat from a running agent process (POST /api/presence) — the
    *  desktop app while open. Null = never seen; stale = offline. */
-  lastSeenAt: z13.string().datetime().nullable().optional(),
+  lastSeenAt: z14.string().datetime().nullable().optional(),
   /** WORKING, NOT JUST CONNECTED (owner, 2026-09-30): the last time the agent itself acted on one of
    *  its Goals — wrote on one or recorded an operation (`tokens.last_worked_at`). Within
    *  `WORKING_MS` it is working; otherwise it is connected but idle. Null = not seen working yet. */
-  lastWorkedAt: z13.string().datetime().nullable().optional(),
+  lastWorkedAt: z14.string().datetime().nullable().optional(),
+  /** A TERMINAL SESSION'S ONLY PULSE (`tokens.listened_at`): `paigy-listen` stamps it on every sweep
+   *  (90 s). It is deliberately NOT presence — `last_seen_at` is the desktop host's beat, and a
+   *  listener orphaned from a dead conversation must not light the live dot — but it is the one fact
+   *  that says a terminal session's process is UP, which is why the wake already reads it
+   *  (`wake.ts`, `listener`). Null = no listener sweeping. */
+  listenedAt: z14.string().datetime().nullable().optional(),
   /** The oldest of its Goals that is `ready` for it — work handed to it that nobody has started.
    *  With no work of its own for `WORKING_MS`, an agent sitting on this is not taking its work. */
-  oldestReadyAt: z13.string().datetime().nullable().optional(),
+  oldestReadyAt: z14.string().datetime().nullable().optional(),
   /** What a live desktop can run (docs/clients/desktop/companion.md §2.2), advertised on its heartbeat:
    *  harness availabilities + granted workspaces — the option set the phone's
    *  "new session" sheet offers. Absent for ordinary MCP agents. */
-  runtime: z13.object({
+  runtime: z14.object({
     /** The @paigy/harness this host is running — a machine the self-update has not reached
      *  shows its age here (`apps/desktop/src/update.ts`). */
-    version: z13.string().optional(),
-    harnesses: z13.array(z13.object({ name: z13.string(), label: z13.string(), status: z13.string(), models: z13.array(z13.string()).optional() })).optional(),
-    workspaces: z13.array(z13.string()).optional(),
+    version: z14.string().optional(),
+    harnesses: z14.array(z14.object({ name: z14.string(), label: z14.string(), status: z14.string(), models: z14.array(z14.string()).optional() })).optional(),
+    workspaces: z14.array(z14.string()).optional(),
     /** THE GIT REPOS IN THOSE FOLDERS (2026-10-01, Goal 26982211): each granted folder that is a
      *  repo, and each repo directly inside one, with its `origin` remote. A session started for
      *  work on `mauurda/paigy` opens in that repo rather than the folder above it, where the repo's
      *  own AGENTS.md is never read (`workspaceForRepo`). Absent on hosts that predate it. */
-    repos: z13.array(z13.object({ path: z13.string(), remote: z13.string() })).optional()
+    repos: z14.array(z14.object({ path: z14.string(), remote: z14.string() })).optional()
   }).optional(),
   /** The tail of this agent's working log, when a harness is driving it — the agent page's
    *  live strip. Absent for anything the desktop harness isn't running (a hatched identity
@@ -1640,151 +1771,151 @@ var ConnectionSummarySchema = z13.object({
   activity: AgentActivitySchema.optional(),
   /** True = a provider-managed agent running in the provider's cloud (e.g. Anthropic CMA);
    *  false = a local MCP connection running on the user's computer (Claude Code/Codex/…). */
-  managed: z13.boolean()
+  managed: z14.boolean()
 });
-var LedgerItemSchema = z13.object({ id: z13.string(), parentId: z13.string(), title: z13.string(), createdAt: z13.string() });
-var AgentLedgerSchema = z13.object({
+var LedgerItemSchema = z14.object({ id: z14.string(), parentId: z14.string(), title: z14.string(), createdAt: z14.string() });
+var AgentLedgerSchema = z14.object({
   /** Null when the agent has not named itself yet — never a placeholder (owner, 2026-10-01). */
-  agent: z13.object({ id: z13.string(), name: z13.string().nullable(), revokedAt: z13.string().nullable() }),
+  agent: z14.object({ id: z14.string(), name: z14.string().nullable(), revokedAt: z14.string().nullable() }),
   /** Its own questions you have not answered. */
-  asks: z13.array(LedgerItemSchema),
+  asks: z14.array(LedgerItemSchema),
   /** Its questions you answered that nobody acted on — still owed to somebody. */
-  answered: z13.array(LedgerItemSchema),
+  answered: z14.array(LedgerItemSchema),
   /** Requests you sent it that it never took. */
-  requests: z13.array(LedgerItemSchema),
-  goals: z13.array(z13.object({ id: z13.string(), outcome: z13.string(), state: z13.string() })),
-  callbacks: z13.array(z13.object({ id: z13.string(), parentId: z13.string(), trigger: z13.string(), note: z13.string(), dueAt: z13.string().nullable() }))
+  requests: z14.array(LedgerItemSchema),
+  goals: z14.array(z14.object({ id: z14.string(), outcome: z14.string(), state: z14.string() })),
+  callbacks: z14.array(z14.object({ id: z14.string(), parentId: z14.string(), trigger: z14.string(), note: z14.string(), dueAt: z14.string().nullable() }))
 });
-var ReassignResultSchema = z13.object({
-  moved: z13.object({ asks: z13.number(), answered: z13.number(), requests: z13.number(), goals: z13.number(), callbacks: z13.number() }),
-  parentId: z13.string().nullable()
+var ReassignResultSchema = z14.object({
+  moved: z14.object({ asks: z14.number(), answered: z14.number(), requests: z14.number(), goals: z14.number(), callbacks: z14.number() }),
+  parentId: z14.string().nullable()
 });
-var LessonStateSchema = z13.enum(["active", "proposed", "retired"]);
-var LessonViewSchema = z13.object({
-  id: z13.string(),
-  text: z13.string(),
+var LessonStateSchema = z14.enum(["active", "proposed", "retired"]);
+var LessonViewSchema = z14.object({
+  id: z14.string(),
+  text: z14.string(),
   state: LessonStateSchema,
   /** The Goal it is scoped to; null = the whole account. */
-  scopeGoalId: z13.string().nullable(),
-  goalTitle: z13.string().nullable(),
-  version: z13.number(),
-  pinned: z13.boolean(),
+  scopeGoalId: z14.string().nullable(),
+  goalTitle: z14.string().nullable(),
+  version: z14.number(),
+  pinned: z14.boolean(),
   /** When the person last wrote its text themselves. */
-  editedAt: z13.string().nullable(),
-  createdAt: z13.string(),
-  updatedAt: z13.string(),
+  editedAt: z14.string().nullable(),
+  createdAt: z14.string(),
+  updatedAt: z14.string(),
   /** The Entries it came from, oldest first; `words` is null when an Entry has none to show (sealed). */
-  sources: z13.array(z13.object({ entryId: z13.string(), words: z13.string().nullable(), at: z13.string() }))
+  sources: z14.array(z14.object({ entryId: z14.string(), words: z14.string().nullable(), at: z14.string() }))
 });
-var QueueQuestionSchema = z13.object({
+var QueueQuestionSchema = z14.object({
   /** The decision need's id — what an answer is accepted against. */
-  id: z13.string(),
+  id: z14.string(),
   /** The words that were asked, from the request Entry that asked them. */
-  question: z13.string(),
+  question: z14.string(),
   /** Where it was asked — which is where the ruling goes (`POST /api/entries`). Null only
    *  for a need whose request Entry is carried by no interactive Delivery, which nothing
    *  can answer. */
-  deliveryId: z13.string().nullable().default(null),
+  deliveryId: z14.string().nullable().default(null),
   /** The Entry the ruling is about. */
-  aboutId: z13.string().nullable().default(null),
+  aboutId: z14.string().nullable().default(null),
   /** Empty for a free-text question. */
-  options: z13.array(OptionSchema).default([]),
-  select: z13.enum(["one", "many", "rank", "confirm", "text"]).default("text"),
-  askedAt: z13.string(),
+  options: z14.array(OptionSchema).default([]),
+  select: z14.enum(["one", "many", "rank", "confirm", "text"]).default("text"),
+  askedAt: z14.string(),
   /** Null while the question is open — which is how the page tells the two apart. */
-  answeredAt: z13.string().nullable().default(null),
+  answeredAt: z14.string().nullable().default(null),
   /** The ruling in the person's own words, from the contribution that replied — not the
    *  option id, which is not something anyone reads back. Null while it is open, and null
    *  for a settled question whose reply carried nothing readable. */
-  answer: z13.string().nullable().default(null),
+  answer: z14.string().nullable().default(null),
   /** The Goal this question belongs to — a step knows its Goal on its own, not only through
    *  an `InboxItem`'s `communication.goalIds[0]` (docs/clients/app/walk/design.md §12 item 3).
    *  READ BY `apps/client/src/walk/order.ts`, which stamps it onto every `WalkStep`: the walk's
    *  order, its route, home's trees and the list of steps all take a step's Goal from here, so
    *  this is the field they agree through rather than each re-deriving it from the row it
    *  arrived under. Required because the API projects it on every need it sends. */
-  goalId: z13.string(),
+  goalId: z14.string(),
   /** True only while an unmet START gate holds the Goal — a Goal that merely waits to
    *  *finish* does not stop a person from answering (owner, 2026-09-16: "per need gate from
    *  the API"; §4's dashed node). Not the same fact as `QueueItem.blocked`, which counts any
    *  gate at all. */
-  blocked: z13.boolean().default(false)
+  blocked: z14.boolean().default(false)
 });
-var QueueReplySchema = z13.object({
+var QueueReplySchema = z14.object({
   /** The card this note was (`deliveryId:requestEntryId`, minted by the server like every card
    *  id) — so the phone can tell a reply it just sent from one the queue already carries, and the
    *  walk can name it in its zoom. */
-  id: z13.string(),
+  id: z14.string(),
   /** The Goal the note is on. */
-  goalId: z13.string(),
+  goalId: z14.string(),
   /** What the note said. */
-  note: z13.string(),
+  note: z14.string(),
   /** Where it was carried — where a second reply goes (`POST /api/entries`, #2252). */
-  deliveryId: z13.string(),
-  requestEntryId: z13.string(),
-  askedAt: z13.string(),
+  deliveryId: z14.string(),
+  requestEntryId: z14.string(),
+  askedAt: z14.string(),
   /** When the person last replied — the window's start. */
-  repliedAt: z13.string(),
+  repliedAt: z14.string(),
   /** The person's latest words about it; null when there is nothing readable in them. */
-  reply: z13.string().nullable()
+  reply: z14.string().nullable()
 });
-var QueueItemSchema = z13.object({
-  id: z13.string(),
+var QueueItemSchema = z14.object({
+  id: z14.string(),
   /** One-line headline — the first sentence of the outcome. */
-  title: z13.string(),
+  title: z14.string(),
   /** The outcome in full, verbatim: the person's own words are what an assignee sees. */
-  intent: z13.string(),
+  intent: z14.string(),
   /** `ready` | `active` | `waiting` | `done` | `cancelled`, straight off the Goal. */
-  state: z13.string(),
+  state: z14.string(),
   /** Who holds it (a participant ref); null when nobody does yet. */
-  assignee: z13.string().nullable().default(null),
+  assignee: z14.string().nullable().default(null),
   /** What the agent last said it was doing; null if it has said nothing. */
-  progress: z13.string().nullable().default(null),
+  progress: z14.string().nullable().default(null),
   /** HOME'S LINE FOR THAT NOTE (owner, 2026-09-23): a few plain words one read wrote from `progress`,
    *  served only while it was written for the current note. Null means show the Goal's name. */
-  progressLine: z13.string().nullable().optional(),
-  reviewPending: z13.boolean().default(false),
-  dueAt: z13.string().nullable().default(null),
+  progressLine: z14.string().nullable().optional(),
+  reviewPending: z14.boolean().default(false),
+  dueAt: z14.string().nullable().default(null),
   /** WHEN ITS OWNER SAID DONE WHILE CHILDREN WERE OPEN (#2704): its own work is finished and it closes
    *  with its last open child. Null otherwise; optional, so hand-built queues need not spell it. */
-  finishedAt: z13.string().nullable().optional(),
+  finishedAt: z14.string().nullable().optional(),
   /** The Goal this one was opened under; null at the root. */
-  parentGoalId: z13.string().nullable().default(null),
+  parentGoalId: z14.string().nullable().default(null),
   /** Goals opened under this one — only those the same list holds. */
-  childGoalIds: z13.array(z13.string()).default([]),
+  childGoalIds: z14.array(z14.string()).default([]),
   /** Goals this one waits on (start or finish gates). */
-  dependencyGoalIds: z13.array(z13.string()).default([]),
+  dependencyGoalIds: z14.array(z14.string()).default([]),
   /** True while any gate is on a Goal that is not done — the walk draws it dashed. */
-  blocked: z13.boolean().default(false),
+  blocked: z14.boolean().default(false),
   /** Its questions: every OPEN one, and at most ten settled, newest settled first
    *  (20260929133308) — the page decides which of them to show. NOT the whole set: `asked` and
    *  `answered` are, and a settled one's words are a line (280 characters), its body read when the
    *  question is opened. Never a CANCELLED one (withdrawn, dismissed, or its Goal or call ended): it
    *  asks nothing of anyone (#3723). */
-  questions: z13.array(QueueQuestionSchema).default([]),
+  questions: z14.array(QueueQuestionSchema).default([]),
   /** HOW MANY QUESTIONS THIS WORK HAS ASKED, and how many are answered — the Goal's own totals,
    *  bounded at 100 server-side. A tally counted off `questions` is a wrong number that looks
    *  right once the cap bites (`walk/trees.ts` `tallyOf`). Optional, and defaulted from the array
    *  by the projection, so hand-built queues (fixtures, the demo) need not spell them. A cancelled
    *  question is not asked (#3725). */
-  asked: z13.number().optional(),
-  answered: z13.number().optional(),
+  asked: z14.number().optional(),
+  answered: z14.number().optional(),
   /** Every note on it the person replied to (`QueueReplySchema`) — the page decides which to show.
    *  Optional, not defaulted: absent is none, and every hand-built queue (fixtures, the demo) need
    *  not spell an empty list. */
-  replies: z13.array(QueueReplySchema).optional(),
+  replies: z14.array(QueueReplySchema).optional(),
   /** The repository or project identifier this Goal belongs to (#2280), null if untracked. */
-  repo: z13.string().nullable().optional(),
-  createdAt: z13.string(),
-  updatedAt: z13.string().nullable().default(null),
+  repo: z14.string().nullable().optional(),
+  createdAt: z14.string(),
+  updatedAt: z14.string().nullable().default(null),
   /** When its owner last SAID something about it (the newest `progress` Entry: a contact update kept
    *  as progress). `updatedAt` moves for reasons nobody chose — a
    *  state recomputed, a review flag — so it cannot tell work in hand from work gone quiet. */
-  lastProgressAt: z13.string().nullable().optional(),
+  lastProgressAt: z14.string().nullable().optional(),
   /** WHEN AN AGENT LAST ACTED ON IT (owner, 2026-10-27: "Move it now", Goal 567258dc): its newest agent write, an
    *  Entry or a Goal change, Paigy's excluded (`goals.worked_at`). What keeps work an agent is writing on from
    *  reading as gone quiet (`coldSince`); absent on a door that does not carry it. */
-  workedAt: z13.string().nullable().optional(),
+  workedAt: z14.string().nullable().optional(),
   /** THE GOAL'S NEWEST WORD, FROM EITHER SIDE (owner, 2026-09-27): the newest Entry on it, of any
    *  kind — what the person added ("Add to this"), their reply, the agent's ask or its progress
    *  note. A progress note is an Entry, so this is already the newer of the two: the person's note
@@ -1803,7 +1934,7 @@ var QueueItemSchema = z13.object({
    *  minute after the person speaks erases their instant from it, and the durable traces the client
    *  can see (`replies`, `questions[].answeredAt`) miss a spontaneous note entirely — a `request`
    *  Entry with no `about_id` is in neither. */
-  lastPersonAt: z13.string().nullable().optional(),
+  lastPersonAt: z14.string().nullable().optional(),
   /** EACH SIDE'S NEWEST WORD IN THE LAST DAY (owner, 2026-10-13, the Work tab's order): the person's
    *  (`yours`) and the agents' (`theirs`), each null when that side wrote nothing on it in 24 hours.
    *  `latest` holds one Entry, so whichever side wrote first is hidden there; Work orders a Goal by
@@ -1811,6 +1942,11 @@ var QueueItemSchema = z13.object({
    *  and a door older than 20261013045542 need not spell them. */
   yours: GoalLatestSchema.nullable().optional(),
   theirs: GoalLatestSchema.nullable().optional(),
+  /** ITS STEPS, COUNTED IN THE DATABASE (owner, 2026-11-04: "Count on the server, and agents make steps"): its
+   *  direct children that were not cancelled, and how many of them are done. The queue holds only a recent
+   *  slice of finished work, so counting `childGoalIds` read "0 of 1" for a project with five done. Absent
+   *  from a door older than this field, or when the count failed: the app then counts what it holds. */
+  steps: z14.object({ total: z14.number().int().nonnegative(), done: z14.number().int().nonnegative() }).nullable().optional(),
   /** WHAT THIS ROW IS, IN TWELVE CHARACTERS (#2928) — the hash of every other field on it, stamped
    *  by the one projection that builds the row (`apps/api/src/goal/queue.ts`). It is how the
    *  incremental read knows a row has not moved: the phone echoes back the revs it holds
@@ -1821,40 +1957,40 @@ var QueueItemSchema = z13.object({
    *  the row shows that no `updated_at` moves for (`active` lapsing, a dependency's state, a
    *  sibling appearing in `childGoalIds`) cannot go unnoticed. Optional because a hand-built
    *  queue (a fixture, the demo) spells none, and a row with no rev is simply always re-sent. */
-  rev: z13.string().optional()
+  rev: z14.string().optional()
 });
-var QueueDeltaSchema = z13.object({
-  ids: z13.array(z13.string()),
-  items: z13.array(QueueItemSchema)
+var QueueDeltaSchema = z14.object({
+  ids: z14.array(z14.string()),
+  items: z14.array(QueueItemSchema)
 });
 var COLD_AFTER_MS = 3 * 24 * 60 * 60 * 1e3;
-var NoteSourceSchema = z13.enum(["app", "call"]);
-var NoteStatusSchema = z13.enum(["open", "assigned", "in_progress", "done"]);
-var NoteRepeatSchema = z13.enum(["once", "until_done"]);
-var DecisionSchema = z13.object({
-  id: z13.string(),
+var NoteSourceSchema = z14.enum(["app", "call"]);
+var NoteStatusSchema = z14.enum(["open", "assigned", "in_progress", "done"]);
+var NoteRepeatSchema = z14.enum(["once", "until_done"]);
+var DecisionSchema = z14.object({
+  id: z14.string(),
   /** The note this decision refines; null = recorded on a bare thread (the
    *  extensibility seam — any conversation can accrue decisions). */
-  noteId: z13.string().nullable(),
+  noteId: z14.string().nullable(),
   /** What was ambiguous — the broker's (or the user's own) question. */
-  question: z13.string(),
+  question: z14.string(),
   /** The user's ruling; null while the question is open. */
-  answer: z13.string().nullable(),
-  decidedAt: z13.string().nullable(),
-  createdAt: z13.string()
+  answer: z14.string().nullable(),
+  decidedAt: z14.string().nullable(),
+  createdAt: z14.string()
 });
-var NoteSchema = z13.object({
-  id: z13.string(),
+var NoteSchema = z14.object({
+  id: z14.string(),
   /** One-line headline (broker-titled; deterministic floor). */
-  title: z13.string(),
+  title: z14.string(),
   /** The original intent, verbatim — assignees always see the user's own words. */
-  intent: z13.string(),
+  intent: z14.string(),
   source: NoteSourceSchema,
   status: NoteStatusSchema,
   /** Who it was assigned to (a participant ref, 'agent:<tokenId>'); null = unassigned. */
-  assignee: z13.string().nullable(),
+  assignee: z14.string().nullable(),
   /** The request thread minted at assignment; null until assigned. */
-  parentId: z13.string().nullable(),
+  parentId: z14.string().nullable(),
   /** REMINDERS (docs/model/notes/reminders-design.md): the NOT-BEFORE this becomes eligible to ride a
    *  call — never a deadline. It only ever comes from the user's own words, so when it
    *  passes Paigy rings ONCE (#1293, owner 2026-08-26: a time said out loud is consent to
@@ -1863,100 +1999,100 @@ var NoteSchema = z13.object({
   //  Defaulted, not required: a Note from an API deploy older than the reminders
   //  migration has none of these, and the defaults ARE what it means — no not-before,
   //  one ride, never ridden. Parsing must not fail across a rolling deploy.
-  dueAt: z13.string().nullable().default(null),
+  dueAt: z14.string().nullable().default(null),
   repeat: NoteRepeatSchema.default("once"),
   /** How many calls have already carried it — the fatigue cap counts rides, not days. */
-  rides: z13.number().int().default(0),
-  lastRideAt: z13.string().nullable().default(null),
-  createdAt: z13.string()
+  rides: z14.number().int().default(0),
+  lastRideAt: z14.string().nullable().default(null),
+  createdAt: z14.string()
 });
-var DeliveryModeSchema = z13.enum(["poll", "self_hosted"]);
-var RegisterDeliverySchema = z13.object({ mode: DeliveryModeSchema });
-var OAuthStartSchema = z13.object({
-  provider: z13.enum(["cma"]),
-  returnTo: z13.string().min(1)
+var DeliveryModeSchema = z14.enum(["poll", "self_hosted"]);
+var RegisterDeliverySchema = z14.object({ mode: DeliveryModeSchema });
+var OAuthStartSchema = z14.object({
+  provider: z14.enum(["cma"]),
+  returnTo: z14.string().min(1)
 });
-var DeliveryConfigSchema = z13.object({
-  tokenId: z13.string(),
+var DeliveryConfigSchema = z14.object({
+  tokenId: z14.string(),
   mode: DeliveryModeSchema,
   /** null when the deployment has no anon key configured. `self_hosted` is then REFUSED
    *  (503 `self_hosted_unavailable`) rather than registered, so a self_hosted config always
    *  carries credentials; only a `poll` registration can come back with null here. */
-  realtime: z13.object({ url: z13.string(), anonKey: z13.string() }).nullable()
+  realtime: z14.object({ url: z14.string(), anonKey: z14.string() }).nullable()
 });
-var HostDecisionSchema = z13.object({
+var HostDecisionSchema = z14.object({
   /** The agent's token id: the row's `recipient`. */
-  agent: z13.string().uuid(),
-  decision: z13.enum(["stood_back", "took_over"]),
+  agent: z14.string().uuid(),
+  decision: z14.enum(["stood_back", "took_over"]),
   /** The work it was about: the Goal waiting on that agent next (`claimable` on its `contact({})` read). */
-  goalId: z13.string().uuid().nullable().optional(),
+  goalId: z14.string().uuid().nullable().optional(),
   /** When the server last heard from the agent, as the host read it: the presence it stood back for. */
-  seenAt: z13.string().datetime().nullable().optional(),
+  seenAt: z14.string().datetime().nullable().optional(),
   /** When that work last moved (`claimable.since` on an agent's `contact({})` read), the fact the bound is judged on. */
-  since: z13.string().datetime().nullable().optional(),
+  since: z14.string().datetime().nullable().optional(),
   /** What the host said, in its log's own words: why it stood back, or what the take-over did. */
-  said: z13.string().max(300).optional()
+  said: z14.string().max(300).optional()
 });
-var WakeNudgeSchema = z13.object({
-  kind: z13.enum(["reply", "request", "callback"]),
-  notificationId: z13.string().optional(),
-  parentId: z13.string()
+var WakeNudgeSchema = z14.object({
+  kind: z14.enum(["reply", "request", "callback"]),
+  notificationId: z14.string().optional(),
+  parentId: z14.string()
 });
-var PairingStatusSchema = z13.enum(["pending", "approved", "denied", "expired"]);
-var DeviceCodeSchema = z13.object({
-  device_code: z13.string(),
-  user_code: z13.string(),
-  verification_uri: z13.string().url(),
-  verification_uri_complete: z13.string().url(),
-  interval: z13.number(),
-  expires_in: z13.number()
+var PairingStatusSchema = z14.enum(["pending", "approved", "denied", "expired"]);
+var DeviceCodeSchema = z14.object({
+  device_code: z14.string(),
+  user_code: z14.string(),
+  verification_uri: z14.string().url(),
+  verification_uri_complete: z14.string().url(),
+  interval: z14.number(),
+  expires_in: z14.number()
 });
-var DeviceInfoSchema = z13.object({
-  code: z13.string(),
+var DeviceInfoSchema = z14.object({
+  code: z14.string(),
   /** The agent's suggested name (from /device/code) — shown on the approval screen,
    *  pre-filling the name field the human can edit. */
-  name: z13.string(),
+  name: z14.string(),
   /** @deprecated Legacy alias of `name` for the pre-#531 embedded bundle in App Store
    *  build 35, whose DeviceFlow renders `info.agent.slice(0, 2)` — without this a FRESH
    *  install crashes on the pairing screen on first launch, before the OTA lands
    *  (seen live: PAIGY-5T, 2026-07-21). Remove once a newer binary is the floor. */
-  agent: z13.string().optional(),
-  device: z13.string().nullable(),
+  agent: z14.string().optional(),
+  device: z14.string().nullable(),
   status: PairingStatusSchema
 });
-var DeviceTokenSchema = z13.object({
-  access_token: z13.string(),
+var DeviceTokenSchema = z14.object({
+  access_token: z14.string(),
   /** The pairing's single name (user-typed at approval, the agent's suggestion, or
    *  a default silly name). */
-  name: z13.string(),
-  device: z13.string().nullable(),
+  name: z14.string(),
+  device: z14.string().nullable(),
   /** The pairing's assigned voice, cached so the desktop can seed the SAME face the phone
    *  draws — voice is the third ingredient of a hatchling's build (party/traits.ts). */
-  voice: z13.string().nullable().optional(),
+  voice: z14.string().nullable().optional(),
   /** The token's server-side id — the face's COLOUR anchor, and the only seed ingredient
    *  that survives a rename. Cached by the host's identity beat. */
-  token_id: z13.string().nullable().optional(),
+  token_id: z14.string().nullable().optional(),
   /** WHERE this identity works — the folder a wake should land it in. Written by the host
    *  at spawn and by `paigy-harness handoff` from a live terminal. Without it every wake
    *  landed in the FIRST granted workspace and the agent rediscovered its own repo from
    *  the thread each time (host.ts, live catch 2026-08-06 — prompt-papered until now). */
-  workspace: z13.string().nullable().optional(),
+  workspace: z14.string().nullable().optional(),
   /** Local host recovery must preserve the launch's runtime and Paigy identity. */
-  harness: z13.enum(["claude", "codex", "agy"]).optional(),
-  session_id: z13.string().uuid().optional(),
+  harness: z14.enum(["claude", "codex", "agy", "grok"]).optional(),
+  session_id: z14.string().uuid().optional(),
   /** A conversation the host must not resume: its context is full, so every turn fails
    *  ("Prompt is too long"). Written when a run hits it (`run.ts` `onFull`); the host skips a slot
    *  whose resumable session is this one, so its Goals reach the dead-agent handoff instead of a
    *  copy that types the person's words into a turn that cannot run (Calls, 2026-10-06). */
-  full_session: z13.string().optional(),
-  uik_pub: z13.string().nullable().optional()
+  full_session: z14.string().optional(),
+  uik_pub: z14.string().nullable().optional()
 });
-var SupportRequestSchema = z13.object({
-  email: z13.string().email().max(320),
-  message: z13.string().trim().min(1).max(5e3),
-  name: z13.string().trim().max(120).optional()
+var SupportRequestSchema = z14.object({
+  email: z14.string().email().max(320),
+  message: z14.string().trim().min(1).max(5e3),
+  name: z14.string().trim().max(120).optional()
 });
-var NotificationFeedbackKindSchema = z13.enum([
+var NotificationFeedbackKindSchema = z14.enum([
   "break_down",
   // "This should be more than one ask — break it down."
   "regenerate_options",
@@ -1971,133 +2107,133 @@ var NotificationFeedbackKindSchema = z13.enum([
   // anything else — the note carries it.
 ]);
 var SlimOptionSchema = OptionSchema.omit({ html: true });
-var QuestionRowSchema = z13.object({
+var QuestionRowSchema = z14.object({
   /** The card's id (`deliveryId:needId`, or `deliveryId:entryId` for an update), as the inbox mints it. */
-  id: z13.string(),
-  deliveryId: z13.string(),
-  entryId: z13.string(),
+  id: z14.string(),
+  deliveryId: z14.string(),
+  entryId: z14.string(),
   /** The decision it waits on; null for an update, which asks nothing. */
-  needId: z13.string().nullable(),
-  goalIds: z13.array(z13.string()),
+  needId: z14.string().nullable(),
+  goalIds: z14.array(z14.string()),
   /** The name of the work it is about, when the read could word it. */
-  goalTitle: z13.string().optional(),
-  tokenId: z13.string().optional(),
-  name: z13.string(),
-  title: z13.string(),
-  body: z13.string(),
-  select: z13.enum(["one", "many", "rank", "confirm", "text"]),
-  options: z13.array(SlimOptionSchema),
-  hasPreview: z13.boolean(),
-  blocking: z13.boolean(),
-  askedAt: z13.string().datetime(),
+  goalTitle: z14.string().optional(),
+  tokenId: z14.string().optional(),
+  name: z14.string(),
+  title: z14.string(),
+  body: z14.string(),
+  select: z14.enum(["one", "many", "rank", "confirm", "text"]),
+  options: z14.array(SlimOptionSchema),
+  hasPreview: z14.boolean(),
+  blocking: z14.boolean(),
+  askedAt: z14.string().datetime(),
   ring: InboxItemSchema.shape.ring,
-  onCall: z13.literal(true).optional(),
-  sealed: z13.boolean()
+  onCall: z14.literal(true).optional(),
+  sealed: z14.boolean()
 });
-var WorkStateSchema = z13.enum(["ready", "active", "waiting", "done", "cancelled"]);
-var WorkRowSchema = z13.object({
-  id: z13.string(),
-  parentId: z13.string().nullable(),
-  title: z13.string(),
+var WorkStateSchema = z14.enum(["ready", "active", "waiting", "done", "cancelled"]);
+var WorkRowSchema = z14.object({
+  id: z14.string(),
+  parentId: z14.string().nullable(),
+  title: z14.string(),
   /** Straight off the Goal. */
   state: WorkStateSchema,
-  owner: z13.string().nullable(),
-  revision: z13.number().int(),
+  owner: z14.string().nullable(),
+  revision: z14.number().int(),
   /** Open questions on it, counted to 100. */
-  waiting: z13.number().int(),
+  waiting: z14.number().int(),
   /** Held by a gate on work that is not done. */
-  blocked: z13.boolean(),
-  lastProgressAt: z13.string().datetime().nullable(),
+  blocked: z14.boolean(),
+  lastProgressAt: z14.string().datetime().nullable(),
   /** The line written for its newest progress note, else that note's first words. */
-  line: z13.string().nullable(),
+  line: z14.string().nullable(),
   /** Work directly under it, counted to 100; the list carries up to 12 of them. */
-  children: z13.number().int(),
-  createdAt: z13.string().datetime(),
-  updatedAt: z13.string().datetime(),
+  children: z14.number().int(),
+  createdAt: z14.string().datetime(),
+  updatedAt: z14.string().datetime(),
   /** When anything at or under it last moved — the order the list is in. */
-  activeAt: z13.string().datetime(),
+  activeAt: z14.string().datetime(),
   /** A sealed outcome has no title here; the work's page opens it. */
-  sealed: z13.boolean()
+  sealed: z14.boolean()
 });
 var ComputerRowSchema = ConnectionSummarySchema.omit({ activity: true });
 var AgentRowSchema = ComputerRowSchema.extend({
   /** Open questions it is asking the person, over every open card; null when that read failed. */
-  asking: z13.number().int().nullable(),
-  oldestAskAt: z13.string().datetime().nullable(),
+  asking: z14.number().int().nullable(),
+  oldestAskAt: z14.string().datetime().nullable(),
   /** Up to three of the live Goals it holds, oldest first (the order it picks them up), and how
    *  many in all among the account's 200 most recently active agent-held live Goals
    *  (`agent_holds`); null when that read failed. */
-  holds: z13.array(z13.object({ id: z13.string(), title: z13.string() })).nullable(),
-  held: z13.number().int().nullable(),
+  holds: z14.array(z14.object({ id: z14.string(), title: z14.string() })).nullable(),
+  held: z14.number().int().nullable(),
   /** The newest line of its working log, and when the harness saw it. */
-  line: z13.string().nullable(),
-  lineAt: z13.string().datetime().nullable()
+  line: z14.string().nullable(),
+  lineAt: z14.string().datetime().nullable()
 });
-var SnapshotSchema = z13.object({
+var SnapshotSchema = z14.object({
   /** The API's clock, taken before the first read: what a later delta will start from. */
-  at: z13.string().datetime(),
-  questions: z13.object({
+  at: z14.string().datetime(),
+  questions: z14.object({
     /** The newest 30 open cards, questions before updates. */
-    items: z13.array(QuestionRowSchema),
+    items: z14.array(QuestionRowSchema),
     /** Every open question, and apart from them every update, and what was put off. */
-    total: z13.number().int(),
-    updates: z13.number().int(),
-    putOff: z13.number().int()
+    total: z14.number().int(),
+    updates: z14.number().int(),
+    putOff: z14.number().int()
   }).nullable(),
-  agents: z13.object({
+  agents: z14.object({
     /** Up to 60, most recently seen first. */
-    items: z13.array(AgentRowSchema),
-    more: z13.boolean()
+    items: z14.array(AgentRowSchema),
+    more: z14.boolean()
   }).nullable(),
-  work: z13.object({
+  work: z14.object({
     /** The 60 most recently active roots, each followed by up to 12 children; 240 rows at most. */
-    items: z13.array(WorkRowSchema),
+    items: z14.array(WorkRowSchema),
     /** How much work is behind each of the Work tab's four filters, each counted to 100, read with
      *  the rows. `work_list` (20260928023533) owns the predicates: Live is `ready`, `active` or
      *  `waiting`; Waiting on you is live work with an open question or an unmet gate; Not started
      *  is `ready`; Done is `done` or `cancelled`. */
-    counts: z13.object({ live: z13.number().int(), waiting: z13.number().int(), notStarted: z13.number().int(), done: z13.number().int() })
+    counts: z14.object({ live: z14.number().int(), waiting: z14.number().int(), notStarted: z14.number().int(), done: z14.number().int() })
   }).nullable(),
-  you: z13.object({
+  you: z14.object({
     settings: UserSettingsSchema,
-    callable: z13.boolean(),
+    callable: z14.boolean(),
     /** Up to 20 paired computers; null when the roster read failed. */
-    computers: z13.array(ComputerRowSchema).nullable()
+    computers: z14.array(ComputerRowSchema).nullable()
   }).nullable()
 });
-var CallRecapSchema = z13.object({
-  call: z13.object({
-    status: z13.string(),
-    startedAt: z13.string(),
-    durationMs: z13.number().nullable(),
+var CallRecapSchema = z14.object({
+  call: z14.object({
+    status: z14.string(),
+    startedAt: z14.string(),
+    durationMs: z14.number().nullable(),
     /** Bot-clock pickup offset; absent before the timestamp release, null before its report lands. */
-    pickupMs: z13.number().nullable().optional(),
+    pickupMs: z14.number().nullable().optional(),
     /** STILL FILING (20261026182117): the call has not ended on the server, or it has and a brain cycle on it is
      *  still pending, the next writer, which may still settle, file or start something. What the close itself
      *  waits for (`private.close_call`). `status` cannot say it: nothing writes "finalizing", and a call reads
      *  "open" until the sweep closes it, after nothing is pending. Absent from an API before it. */
-    filing: z13.boolean().optional(),
-    agents: z13.array(z13.object({ id: z13.string(), name: z13.string().nullable() }))
+    filing: z14.boolean().optional(),
+    agents: z14.array(z14.object({ id: z14.string(), name: z14.string().nullable() }))
   }),
-  topics: z13.array(z13.object({
-    goalId: z13.string().uuid(),
-    title: z13.string(),
-    owner: z13.string(),
-    state: z13.string(),
+  topics: z14.array(z14.object({
+    goalId: z14.string().uuid(),
+    title: z14.string(),
+    owner: z14.string(),
+    state: z14.string(),
     /** Presented Entry content and its author, separately from the current Goal owner. */
-    items: z13.array(z13.object({ entryId: z13.string(), author: z13.string(), text: z13.string(), summary: z13.string().nullable(), acknowledged: z13.boolean().optional() })).optional(),
+    items: z14.array(z14.object({ entryId: z14.string(), author: z14.string(), text: z14.string(), summary: z14.string().nullable(), acknowledged: z14.boolean().optional() })).optional(),
     /** Whether the call put it to the person. An item it carried and never said has no row on the
      *  review (owner, 2026-10-08: "why do we get the third one when it didn't happen in the call?"). */
-    presented: z13.boolean(),
+    presented: z14.boolean(),
     /** Whether THIS call created it — the third way a call touches work, beside carrying a request
      *  there and filing words there (owner, 2026-10-08, Goal `5b4b1a0a`: "Show the new work, keep my
      *  words under it"). Created work is never carried, so it is never `presented`; without this the
      *  row is dropped by the "only what happened on the call" filter. False on a database before
      *  20261013080133. */
-    created: z13.boolean(),
+    created: z14.boolean(),
     /** `request` is the Entry that asked it — what a line's `about` names when the line answered or put
      *  it off. Optional: an older API sent none. */
-    questions: z13.array(z13.object({ id: z13.string().uuid(), state: z13.string(), title: z13.string(), request: z13.string().optional() })),
+    questions: z14.array(z14.object({ id: z14.string().uuid(), state: z14.string(), title: z14.string(), request: z14.string().optional() })),
     /** `words` is always what they SAID, verbatim — the record, never replaced. `summary` is what
      *  the line says in ten words (owner, 2026-10-07): the filer's, or the talker's for the answer it
      *  gave, so the row scans like a chosen option and their own words stay under it. Absent when
@@ -2107,34 +2243,34 @@ var CallRecapSchema = z13.object({
     /** `reply` is the reply the line makes up (an answer, an ok, a deferral), so one reply is one row
      *  however many lines it cites; `at` is the call lines it came from, so a filed line an answer already
      *  said is not drawn again (owner, 2026-10-07: "fix them all"); `replyKind` says which replies are answers. */
-    lines: z13.array(z13.object({
-      entryId: z13.string().uuid(),
-      words: z13.string(),
-      summary: z13.string().optional(),
-      about: z13.string().nullable().optional(),
-      reply: z13.string().nullable().optional(),
-      replyKind: z13.string().nullable().optional(),
-      at: z13.array(z13.number()).optional()
+    lines: z14.array(z14.object({
+      entryId: z14.string().uuid(),
+      words: z14.string(),
+      summary: z14.string().optional(),
+      about: z14.string().nullable().optional(),
+      reply: z14.string().nullable().optional(),
+      replyKind: z14.string().nullable().optional(),
+      at: z14.array(z14.number()).optional()
     }))
   })),
-  unfiled: z13.array(z13.object({ lineId: z13.string().uuid(), words: z13.string(), atMs: z13.number() })),
+  unfiled: z14.array(z14.object({ lineId: z14.string().uuid(), words: z14.string(), atMs: z14.number() })),
   /** THE PUBLIC TRANSCRIPT — every line said on the call, in the order it was said, and WHO SAID IT
    *  (owner, 2026-10-08: "the public view should only have the lines and who said them, and that's
    *  it"; `docs/brain/phase-three/modular.md` build step 3). `speaker` is the role: "user" is the
    *  person, "agent" an agent speaking, named in `agent`. The timeline's action lines are ours, not
    *  theirs, and are not here; nor is a sealed line, which the server has no words for. It stands
    *  BESIDE `topics` and `unfiled` — what the call did with the words — never instead of them. */
-  said: z13.array(z13.object({
-    lineId: z13.string().uuid(),
-    speaker: z13.enum(["user", "agent"]),
-    agent: z13.object({ id: z13.string().nullable(), name: z13.string().nullable() }).nullable(),
-    words: z13.string(),
-    atMs: z13.number()
+  said: z14.array(z14.object({
+    lineId: z14.string().uuid(),
+    speaker: z14.enum(["user", "agent"]),
+    agent: z14.object({ id: z14.string().nullable(), name: z14.string().nullable() }).nullable(),
+    words: z14.string(),
+    atMs: z14.number()
   })),
   /** What did not fit, counted. `settles` counts items past the recap's 100; absent before 20261026165518. */
-  more: z13.object({ lines: z13.number(), entries: z13.number(), topics: z13.number(), settles: z13.number().optional() }),
+  more: z14.object({ lines: z14.number(), entries: z14.number(), topics: z14.number(), settles: z14.number().optional() }),
   /** Last stored position on this page when another exists; absent before paging shipped. */
-  next: z13.number().int().nonnegative().nullable().optional(),
+  next: z14.number().int().nonnegative().nullable().optional(),
   /** THE CALL AS ONE LIST OF ROWS (owner, 2026-10-14: "we only really want a thing like transcript but
    *  built from the summaries", then "the transcript tabs should show the same but with full words").
    *  Both tabs draw these rows, in this order: Summary from `summary`, Transcript from `words`.
@@ -2145,32 +2281,38 @@ var CallRecapSchema = z13.object({
    *  `by` is the row's real author, never inferred from `kind`: a relay Paigy sent to an agent is
    *  Paigy's, not the person's. Optional and additive (owner: "make sure whatever you do is backward
    *  compatible"): an API before it sends none, and the screen falls back to `topics` and `said`. */
-  timeline: z13.array(z13.object({
-    id: z13.string(),
-    position: z13.number(),
-    atMs: z13.number(),
-    kind: z13.enum(["put", "said", "did"]),
-    by: z13.object({ participant: z13.string().nullable(), agentId: z13.string().nullable(), name: z13.string().nullable() }),
-    goalId: z13.string().uuid().nullable(),
-    requestId: z13.string().nullable(),
-    summary: z13.string().nullable(),
-    words: z13.string(),
+  timeline: z14.array(z14.object({
+    id: z14.string(),
+    position: z14.number(),
+    atMs: z14.number(),
+    kind: z14.enum(["put", "said", "did"]),
+    by: z14.object({ participant: z14.string().nullable(), agentId: z14.string().nullable(), name: z14.string().nullable() }),
+    goalId: z14.string().uuid().nullable(),
+    requestId: z14.string().nullable(),
+    summary: z14.string().nullable(),
+    words: z14.string(),
     /** Paigy's sentences folded into this row: on a `put` row, what it said putting the item to the
      *  person (before their next row); on a `said` row, what it said in reply (until their next row).
      *  `atMs` is when each was said; optional, absent before 2026-10-14's fix. */
-    spoken: z13.array(z13.object({ lineId: z13.string(), words: z13.string(), atMs: z13.number().optional() })),
+    spoken: z14.array(z14.object({ lineId: z14.string(), words: z14.string(), atMs: z14.number().optional() })),
     /** On a `said` row: the work these words were filed on, and whether they answered (recorded) a
      *  question. They replace the "Filed on ..." and "Recorded: ..." rows that repeated the person's own
      *  summary (Brain, 2026-10-14). Optional: absent from an API before it. */
-    filedOn: z13.array(z13.object({ goalId: z13.string().uuid(), title: z13.string() })).optional(),
-    recorded: z13.boolean().optional(),
-    settled: z13.object({ requestId: z13.string(), answer: z13.string() }).nullable().optional(),
-    acknowledged: z13.boolean().optional(),
+    filedOn: z14.array(z14.object({ goalId: z14.string().uuid(), title: z14.string() })).optional(),
+    recorded: z14.boolean().optional(),
+    /** On a `said` row: the agents the call SENT those words to, from the sending action's own
+     *  `about.agents` (20261106220220). The review's receipt is drawn from this and from the row's
+     *  own `requestId`, never from the asker of a question the words happened to sit under: on call
+     *  f9670f67 that asker was Interface while the send named Comms. `[]` or absent means nothing
+     *  was sent for that row, which is why the receipt is positive-only. */
+    sentTo: z14.array(z14.object({ id: z14.string().nullable(), name: z14.string() })).optional(),
+    settled: z14.object({ requestId: z14.string(), answer: z14.string() }).nullable().optional(),
+    acknowledged: z14.boolean().optional(),
     /** On a `put` row: still waiting for the person's answer (not settled, not answered, not put off).
      *  The server's judgement; the screen draws "Still waiting for your answer" from it. Optional: absent
      *  from an API before it, where the screen falls back to its own reading. */
-    waiting: z13.boolean().optional(),
-    lineIds: z13.array(z13.string())
+    waiting: z14.boolean().optional(),
+    lineIds: z14.array(z14.string())
   })).optional(),
   /** THE RECAP (owner, 2026-10-26: "the transcript should just be the entries that settled or
    *  acknowledged the items; that's it"). Each item the person's words settled on this call, in the
@@ -2182,13 +2324,13 @@ var CallRecapSchema = z13.object({
    *  with `requestId` null and the Goal's name as `title`; a started Goal may have no `lines` yet. `kind`
    *  is a string, not an enum, so a later kind never fails an older build's parse; `requestId` is the
    *  request the words settled. On the first page only; absent from an API before it. */
-  settles: z13.array(z13.object({
-    requestId: z13.string().nullable(),
-    kind: z13.string(),
-    by: z13.object({ participant: z13.string(), agentId: z13.string().nullable(), name: z13.string() }),
-    goalId: z13.string().uuid().nullable(),
-    title: z13.string(),
-    lines: z13.array(z13.object({ lineId: z13.string(), words: z13.string(), atMs: z13.number() }))
+  settles: z14.array(z14.object({
+    requestId: z14.string().nullable(),
+    kind: z14.string(),
+    by: z14.object({ participant: z14.string(), agentId: z14.string().nullable(), name: z14.string() }),
+    goalId: z14.string().uuid().nullable(),
+    title: z14.string(),
+    lines: z14.array(z14.object({ lineId: z14.string(), words: z14.string(), atMs: z14.number() }))
   })).optional()
 });
 
@@ -2229,6 +2371,7 @@ PAIGY
 exit 0
 `;
 }
+var GROK_STALLED_HOOK = "npx -y -p @paigy/mcp@latest paigy-stalled";
 function withSessionStartHook(existing, command) {
   let config = {};
   if (existing) {
@@ -2244,6 +2387,16 @@ function withSessionStartHook(existing, command) {
   if (!already) starts.push({ hooks: [{ type: "command", command }] });
   return JSON.stringify(config, null, 2);
 }
+function withGrokPaigyHooks(existing, sessionStart) {
+  const withStart = withSessionStartHook(existing, sessionStart);
+  if (withStart === null) return null;
+  const config = JSON.parse(withStart);
+  const hooks = config.hooks ??= {};
+  const stops = hooks.Stop ??= [];
+  const already = stops.some((group) => (group.hooks ?? []).some((h) => (h.command ?? "").includes("paigy-stalled")));
+  if (!already) stops.push({ hooks: [{ type: "command", command: GROK_STALLED_HOOK, timeout: 60 }] });
+  return JSON.stringify(config, null, 2);
+}
 var word = (s) => `'${s.replace(/'/g, "'\\''")}'`;
 function decideListen(f) {
   if (harnessDriven(f.env)) {
@@ -2368,7 +2521,7 @@ function withReach(shaped, reach) {
   if (!reach || !shaped || typeof shaped !== "object" || Array.isArray(shaped)) return shaped;
   return { ...reach, ...shaped };
 }
-var HOOKED_HARNESSES = "Claude Code and Codex";
+var HOOKED_HARNESSES = "Claude Code, Codex, and Grok";
 function listenBlock(decision, f, stale) {
   const block = {
     listen: decision,
@@ -2456,16 +2609,21 @@ function waitingSummary(received) {
 
 export {
   mcpInputSchema,
+  SKILL_MARK,
   AGENT_PROMPTS,
   promptResult,
+  READS,
+  SENDS,
   AGENT_TOOLS,
   LISTEN_COMMAND,
   serverInstructions,
+  SKILL,
   entryWords,
   entryUnit,
   harnessDriven,
   sessionStartHook,
   withSessionStartHook,
+  withGrokPaigyHooks,
   decideListen,
   writeListenMark,
   removeListenMark,