@rallycry/conveyor-agent 11.0.27 → 11.0.28

package/dist/{chunk-7HK6OIV4.js → chunk-ND5INHM4.js}

@@ -47,13 +47,14 @@ import {
   Lifecycle,
   MAX_CI_WAIT_TIMEOUT_MINUTES,
   MAX_FILE_SIZE_BYTES,
+  MAX_MANUAL_TESTS_PER_TASK,
+  MAX_MANUAL_TEST_TITLE_LENGTH,
   PM_CHAT_HISTORY_LIMIT,
   PRE_BUILD_TASK_STATUSES,
   PTY_STREAM_PORT_ATTEMPTS,
   PTY_STREAM_PORT_BASE,
   PortDiscovery,
   PtyStreamFrameReader,
-  ReviewGuideContentSchema,
   SEVERITY_ENUM,
   TAG_DESCRIPTION_MAX,
   TAG_OVERVIEW_MAX,
@@ -63,12 +64,14 @@ import {
   awaitGitReady,
   clearForceFreshCooldown,
   createServiceLogger,
+  describeManualTestRejection,
   encodePtyStreamFrame,
   ensureOnTaskBranch,
   flushAllPendingWork,
   flushPendingChanges,
   forceFreshCooldownNotice,
   forceFreshMintBlocked,
+  formatEvaluateQuestionsResult,
   getCurrentBranch,
   git,
   hasTaskPlan,
@@ -92,7 +95,7 @@ import {
   statWorkspacePath,
   updateRemoteToken,
   verifyGitCredential
-} from "./chunk-PBSFLQA7.js";
+} from "./chunk-GHNYP67H.js";
 import {
   registerBootMilestoneSocketFallback,
   reportBootMilestone
@@ -6748,6 +6751,7 @@ function compileNumber(z11, spec) {
 function compileArray(z11, spec) {
   let schema = z11.array(compileField(z11, spec.item));
   if (spec.min !== void 0) schema = schema.min(spec.min);
+  if (spec.max !== void 0) schema = schema.max(spec.max);
   return schema;
 }
 function compileBase(z11, spec) {
@@ -6877,16 +6881,24 @@ var readTaskChatContract = defineToolContract({
     }
   }
 });
+var listTagsDetail = f.optional(
+  f.enum(["summary", "full"], {
+    desc: `"summary" (default) returns one compact row per tag with a contextPathCount. "full" also ships each tag's contextPaths inline (each link with its verification status). Full mode can exceed a client's tool-output budget on a large glossary; use get_tag for one term instead.`
+  })
+);
 var listTagsContract = defineToolContract({
   name: "list_tags",
   agent: {
-    description: "List the project glossary: every tag's id, name, color, description, parent/child tag ids, its contextPaths (the rule/doc/file/folder links it wires into agent context \u2014 `[]` means none), whether it carries a full overview (fetch that with get_tag), and its attachmentCount (files labelled as examples of the term). The context links ship inline, so you only need get_tag for a term's full overview. Use the ids with get_tag / update_tag.",
-    fields: {}
+    description: "List the whole project glossary in one call. The default summary gives every tag's id, name, color, description, parent/child tag ids, whether it carries a full overview (hasOverview), its attachmentCount (files labelled as examples of the term), and its contextPathCount (the rule/doc/file/folder links it wires into agent context \u2014 `0` means none). Pass detail: \"full\" to ship the contextPaths themselves inline. Use get_tag for one term's full overview and full link verification. Use the ids with get_tag / update_tag.",
+    fields: {
+      detail: listTagsDetail
+    }
   },
   mcp: {
-    description: "List all project tags with names, IDs, colors, descriptions, hierarchy (parent/child ids), contextPaths (the rule/doc/file/folder links each tag wires into agent context \u2014 `[]` means none), a hasOverview flag, and an attachmentCount (files labelled as examples of the term \u2014 read the tiles with list_tag_attachments). Context links ship inline; call get_tag only for a term's full overview. Pass projectId to target a specific project; otherwise the configured default project is used.",
+    description: 'List all project tags in one call. The default summary gives names, IDs, colors, descriptions, hierarchy (parent/child ids), a hasOverview flag, an attachmentCount (files labelled as examples of the term \u2014 read the tiles with list_tag_attachments), and a contextPathCount (the rule/doc/file/folder links each tag wires into agent context \u2014 `0` means none). Pass detail: "full" to ship the contextPaths themselves inline. Call get_tag for one term\'s full overview and full link verification. Pass projectId to target a specific project; otherwise the configured default project is used.',
     fields: {
-      projectId: mcpProjectId
+      projectId: mcpProjectId,
+      detail: listTagsDetail
     }
   }
 });
@@ -7022,7 +7034,7 @@ var updateTaskContract = defineToolContract({
     }
   },
   mcp: {
-    description: "Update task fields: title, description, plan, status, risk, story points, assignment, githubBranch, or tags. Set status to claim a card (InProgress), triage it (Open), or cancel it (Cancelled); for review approvals prefer approve_task / request_changes, which guard against stale-state races. Tags are additive/subtractive \u2014 pass addTags/removeTags with tag names (not a replace-set). Pass projectId to target a specific project; otherwise the configured default project is used. Moving a task beyond Planning auto-fills any missing icon, story points, and agent assignment \u2014 don't spend turns on them; pass storyPointValue only to correct the sizing yourself. For subtasks use update_subtask.",
+    description: "Update task fields: title, description, plan, status, risk, story points, assignment, githubBranch, onHold, or tags. Set status to claim a card (InProgress), triage it (Open), or cancel it (Cancelled); for review approvals prefer approve_task / request_changes, which guard against stale-state races. Set onHold to park a card (true) or bring it back to the active board (false) without cancelling it. Tags are additive/subtractive \u2014 pass addTags/removeTags with tag names (not a replace-set). Pass projectId to target a specific project; otherwise the configured default project is used. Moving a task beyond Planning auto-fills any missing icon, story points, and agent assignment \u2014 don't spend turns on them; pass storyPointValue only to correct the sizing yourself. For subtasks use update_subtask.",
     fields: {
       projectId: mcpProjectId,
       taskId: f.string({ desc: "The task ID" }),
@@ -7053,6 +7065,11 @@ var updateTaskContract = defineToolContract({
         )
       ),
       githubBranch: f.optional(f.nullable(f.string({ desc: GITHUB_BRANCH_DESC }))),
+      onHold: f.optional(
+        f.boolean({
+          desc: "Park the card on hold (true) or return it to the active board (false). On hold hides the card from the active columns without cancelling it; a card already at an active work status is demoted to Open when parked."
+        })
+      ),
       addTags: f.optional(
         f.array(f.string(), {
           desc: 'Tag names to add to the card (e.g. ["refactor"]). Additive \u2014 existing tags are kept. Unknown names are rejected; use list_tags to see available tags or manage_tags to create one.'
@@ -7096,9 +7113,20 @@ var testStatuses = f.optional(
   })
 );
 var setManualTestItems = f.array(
-  f.object({ title: f.string({ min: 1, desc: "A concise, actionable test step" }) }),
-  { min: 1, desc: "List of manual test steps to add" }
+  f.object({
+    title: f.string({
+      min: 1,
+      max: MAX_MANUAL_TEST_TITLE_LENGTH,
+      desc: "One sentence a non-engineer can follow: where to go, what to do, what they should see. No commands, no code."
+    })
+  }),
+  {
+    min: 1,
+    max: MAX_MANUAL_TESTS_PER_TASK,
+    desc: `Manual tests to add. A card holds at most ${MAX_MANUAL_TESTS_PER_TASK}; most changes need 1, and a change with no user-visible behavior needs none.`
+  }
 );
+var setManualTestsBar = 'A manual test is a real user path in plain language ("Open a card with a PR: the Pull Request slat sits directly under Description"), not an engineering check. Never record lint, typecheck, unit tests, CI, curl, database, log, or deploy steps; those run automatically. Record at most 5 per card, usually 1, and none when the change has no user-visible behavior. Existing items with the same title are skipped; items that name an automated step or exceed the cap are rejected and reported back.';
 var titleToEdit = f.string({ min: 1, desc: "The current title of the manual test to edit" });
 var newTitle = f.string({ min: 1, desc: "The new title for the manual test" });
 var titleToRemove = f.string({ min: 1, desc: "The title of the manual test to remove" });
@@ -7152,13 +7180,13 @@ var queryManualTestsContract = defineToolContract({
 var setManualTestsContract = defineToolContract({
   name: "set_manual_tests",
   agent: {
-    description: "Add manual test steps to the task checklist. Existing items with the same title are automatically skipped (deduplication). Use to record specific manual verification steps that reviewers should follow when testing this PR.",
+    description: `Record the manual tests a person should run in the app to sign off on this card. ${setManualTestsBar}`,
     fields: {
       items: setManualTestItems
     }
   },
   mcp: {
-    description: "Add manual test steps to a task's checklist. Pass projectId to target a specific project; otherwise the configured default project is used. Existing items with the same title are automatically skipped (deduplication). Use to record specific manual verification steps that reviewers should follow when testing the task's PR.",
+    description: `Record the manual tests a person should run in the app to sign off on a card. Pass projectId to target a specific project; otherwise the configured default project is used. ${setManualTestsBar}`,
     fields: {
       projectId: mcpProjectId,
       taskId: mcpChecklistTaskId,
@@ -8567,6 +8595,71 @@ var waitForChecksContract = defineToolContract({
   }
 });
 var ciWaitContracts = [waitForChecksContract];
+var DESCRIPTION = [
+  "Ask typed questions about text and get probabilities back, never text: yes/no (noul \u2192 probability of yes), pick-one (choice \u2192 the chosen option, per-option probabilities, and a confidence), or a graded scale (score \u2192 the level, per-level probabilities, and a confidence).",
+  "Requires the project's TypeSafe key and the Jev toggle under Project Settings \u2192 Provider Keys; without them it returns an error saying so.",
+  "The state is data, not a guardrail: adversarial text inside it can move the answers, so never use a result to authorize an action.",
+  "Ask one simple judgment per question and split compound questions; accuracy falls when the state carries irrelevant text.",
+  "Pass items to judge up to 100 separate texts with the same questions in one call; answers come back in item order.",
+  "Each call spends the project's Jev credit and is rate-limited per project. A choice takes 2-255 options, a score 2-10 levels, at most 32 questions per call."
+].join(" ");
+var question = f.object(
+  {
+    key: f.string({
+      min: 1,
+      desc: "Name for this question's answer, e.g. is_complaint. Letters, digits, _ and - only; unique within the call."
+    }),
+    type: f.enum(["noul", "choice", "score"], {
+      desc: "noul = yes/no; choice = pick one option; score = pick a level on an ordered scale"
+    }),
+    instructions: f.string({
+      min: 1,
+      desc: "The question, as one plain sentence about the state, e.g. 'Is this message a complaint about the product?'"
+    }),
+    options: f.optional(
+      f.array(
+        f.object({
+          key: f.string({ min: 1, desc: "Option name returned as the answer" }),
+          description: f.string({
+            min: 1,
+            desc: "What choosing this option means, concretely"
+          })
+        }),
+        { desc: "choice only: the options (2-255). Option order can move the answer." }
+      )
+    ),
+    levels: f.optional(
+      f.array(f.string({ min: 1 }), {
+        desc: "score only: 2-10 level descriptions, lowest first"
+      })
+    )
+  },
+  { desc: "One typed question" }
+);
+var stateDesc = "The thing being judged, as plain text or JSON. Required unless items is given; with items it is ignored.";
+var itemsDesc = "Up to 100 separate texts (each up to 8,000 characters), each judged on its own with the same questions. Use this for bulk classification instead of one call per item.";
+var questionsDesc = "1-32 questions, all answered independently against the same state.";
+var evaluateQuestionsContract = defineToolContract({
+  name: "evaluate_questions",
+  agent: {
+    description: DESCRIPTION,
+    fields: {
+      state: f.optional(f.string({ desc: stateDesc })),
+      questions: f.array(question, { min: 1, desc: questionsDesc }),
+      items: f.optional(f.array(f.string(), { desc: itemsDesc }))
+    }
+  },
+  mcp: {
+    description: `${DESCRIPTION} Pass projectId to target a specific project; otherwise the configured default project is used.`,
+    fields: {
+      projectId: mcpProjectId,
+      state: f.optional(f.string({ desc: stateDesc })),
+      questions: f.array(question, { min: 1, desc: questionsDesc }),
+      items: f.optional(f.array(f.string(), { desc: itemsDesc }))
+    }
+  }
+});
+var evaluateContracts = [evaluateQuestionsContract];
 var TOOL_CONTRACTS = Object.fromEntries(
   [
     ...tasksContracts,
@@ -8583,7 +8676,8 @@ var TOOL_CONTRACTS = Object.fromEntries(
     ...driveContracts,
     ...meetingsContracts,
     ...logsContracts,
-    ...ciWaitContracts
+    ...ciWaitContracts,
+    ...evaluateContracts
   ].map((contract) => [contract.name, contract])
 );
 
@@ -9588,6 +9682,7 @@ function buildSetManualTestsTool(connection) {
       });
       const parts = [`Created ${result.created} manual test item(s).`];
       if (result.skipped > 0) parts.push(`Skipped ${result.skipped} duplicate(s).`);
+      for (const item of result.rejected ?? []) parts.push(describeManualTestRejection(item));
       return textResult(parts.join(" "));
     } catch (error) {
       const msg = error instanceof Error ? error.message : "Unknown error";
@@ -10149,9 +10244,9 @@ async function rejectBadContextPaths(contextPaths, workspaceDir) {
 function buildListTagsTool(connection, projectId) {
   return defineContractTool(
     listTagsContract,
-    async () => {
+    async ({ detail }) => {
       try {
-        const tags = await connection.call("listProjectTags", { projectId });
+        const tags = await connection.call("listProjectTags", { projectId, detail });
         return textResult(JSON.stringify(tags, null, 2));
       } catch (error) {
         return errText("Failed to list tags", error);
@@ -11016,6 +11111,31 @@ function settleTools(connection, projectId) {
   ];
 }
 
+// src/tools/evaluate-tools.ts
+function buildEvaluateTools(connection, projectId) {
+  return [
+    defineContractTool(
+      evaluateQuestionsContract,
+      async (input) => {
+        try {
+          const res = await connection.call("evaluateQuestions", {
+            projectId,
+            state: input.state,
+            items: input.items,
+            questions: input.questions
+          });
+          return textResult(formatEvaluateQuestionsResult(res));
+        } catch (error) {
+          return textResult(
+            `evaluate_questions failed: ${error instanceof Error ? error.message : "Unknown error"}`
+          );
+        }
+      },
+      { annotations: { readOnlyHint: true } }
+    )
+  ];
+}
+
 // src/tools/connected-tools.ts
 function connectedToolsFor(connection, context) {
   const projectId = context?.projectId;
@@ -11028,6 +11148,9 @@ function connectedToolsFor(connection, context) {
     // Always available: the decision tables exist on every project, and a fork
     // an agent cannot file is a fork it assumes its way past instead.
     ...buildDecisionTools(connection, projectId),
+    // Ungated: whether the project has Jev on is answered by calling it, and
+    // the error names what an admin has to switch on.
+    ...buildEvaluateTools(connection, projectId),
     ...context.googleAnalyticsConfigured ? [buildAnalyticsSummaryTool(connection, projectId)] : [],
     ...context.gcpLogsConfigured ? [buildQueryGcpLogsTool(connection, projectId)] : [],
     ...context.grafanaLogsConfigured ? [buildQueryGrafanaLogsTool(connection, projectId)] : []
@@ -11035,8 +11158,6 @@ function connectedToolsFor(connection, context) {
 }
 
 // src/tools/code-review-tools.ts
-import { execFile } from "child_process";
-import { promisify } from "util";
 import { z as z10 } from "zod";
 async function endReviewSession(connection, reason) {
   await connection.call("endReviewSession", {
@@ -11047,101 +11168,6 @@ async function endReviewSession(connection, reason) {
 var RISK_LEVELS = ["critical", "high", "medium", "low"];
 var reviewedShaSchema = z10.string().regex(/^[0-9a-f]{40}$/i).describe("REQUIRED. The full 40-character commit SHA this verdict reviews.");
 var riskDescription = "REQUIRED. The risk level this change carries, judged by the surface area it touches: critical = touches critical/foundational surface, high = important surface, medium = moderate, low = small/isolated. Set this on every verdict. You have authority to override a risk level already set on the task if you disagree with it.";
-var ReviewGuideToolSchema = z10.strictObject({
-  reviewedSha: z10.string().regex(/^[0-9a-f]{40}$/i).describe(
-    "REQUIRED. The PR's current head as a full 40-char SHA. Run `git rev-parse HEAD` immediately before this call \u2014 never extend an abbreviated hash into 40 characters."
-  ),
-  overview: z10.string().min(1).max(6e4).describe("REQUIRED. Plain-text walkthrough intro, max 3000 characters. Keep it short."),
-  sections: z10.array(
-    z10.strictObject({
-      title: z10.string().min(1).max(160),
-      explanation: z10.string().min(1).max(2e3),
-      classification: z10.enum(["core", "supporting"]).optional(),
-      files: z10.array(
-        z10.strictObject({
-          path: z10.string().min(1).max(500).describe(
-            "A file the PR's diff actually changed. Context files you merely read are rejected."
-          ),
-          startLine: z10.number().int().positive().max(1e6).optional(),
-          endLine: z10.number().int().positive().max(1e6).optional(),
-          hunkHeader: z10.string().min(1).max(300).optional().describe(
-            "Optional anchor, matched byte-exactly against the full hunk header line from `git diff` INCLUDING the context text after the second @@. Copy it verbatim from `git diff <base>..HEAD -- <file> | grep '^@@'`, or omit anchors entirely (path-only entries always validate)."
-          )
-        })
-      ).min(1).max(20)
-    })
-  ).min(1).max(12).optional().describe(
-    "REQUIRED top-level array (never text appended to overview) of ordered conceptual sections."
-  )
-});
-var FLATTENED_SECTIONS_PATTERN = /<\/overview>\s*(?:<parameter name="sections">|<sections>)/;
-var TRAILING_TAGS_PATTERN = /(?:\s*<\/(?:parameter|sections|invoke)>)+\s*$/;
-function sliceOutermostArray(tail) {
-  const start = tail.indexOf("[");
-  const end = tail.lastIndexOf("]");
-  return start >= 0 && end > start ? tail.slice(start, end + 1) : null;
-}
-function recoverFlattenedGuide(overview) {
-  const match = FLATTENED_SECTIONS_PATTERN.exec(overview);
-  if (!match) return null;
-  const head = overview.slice(0, match.index).trim();
-  const tail = overview.slice(match.index + match[0].length).replace(TRAILING_TAGS_PATTERN, "").trim();
-  for (const candidate of [tail, sliceOutermostArray(tail)]) {
-    if (!candidate) continue;
-    try {
-      return { overview: head, sections: JSON.parse(candidate) };
-    } catch {
-    }
-  }
-  return null;
-}
-async function resolveGitHeadSha(cwd) {
-  try {
-    const stdout = workbenchEnabled() ? (await getWorkbenchClient().execFile("git", ["rev-parse", "HEAD"], {
-      cwd,
-      timeout: 1e4
-    })).stdout : (await promisify(execFile)("git", ["rev-parse", "HEAD"], { cwd, timeout: 1e4 })).stdout;
-    const sha = stdout.trim();
-    return /^[0-9a-f]{40}$/i.test(sha) ? sha : null;
-  } catch {
-    return null;
-  }
-}
-function buildPublishReviewGuideTool(connection, options = {}) {
-  const { resolveHeadSha } = options;
-  return defineTool(
-    "publish_review_guide",
-    "Publish or update the PR guide for the CURRENT PR head commit. Call right after create_pull_request succeeds, and again after every push to the branch. The head SHA is resolved from the local repo, so never hand-expand a short hash. Best-effort \u2014 it does not gate opening or updating the PR.",
-    ReviewGuideToolSchema.shape,
-    async ({ reviewedSha, overview, sections }) => {
-      let resolvedOverview = overview;
-      let resolvedSections = sections;
-      if (resolvedSections === void 0) {
-        const recovered = recoverFlattenedGuide(overview);
-        if (!recovered) {
-          throw new Error(
-            "publish_review_guide requires a top-level `sections` array \u2014 an ordered list of conceptual sections, each with title, explanation, and files[]. Your call arrived with only `reviewedSha` and `overview`; the `sections` parameter never reached the server, which usually means the call encoding flattened it into `overview`. Do not append the sections to `overview` as text. Retry with sections as a real array argument, and shorten `overview` if the failure repeats."
-          );
-        }
-        resolvedOverview = recovered.overview;
-        resolvedSections = recovered.sections;
-      }
-      const content = ReviewGuideContentSchema.parse({
-        overview: resolvedOverview,
-        sections: resolvedSections
-      });
-      const result = await connection.call("publishReviewGuide", {
-        sessionId: connection.sessionId,
-        reviewedSha: (resolveHeadSha ? await resolveHeadSha() : null) ?? reviewedSha,
-        ...content
-      });
-      return textResult(
-        `Review guide published for ${result.reviewedSha}${result.replaced ? " (replaced)" : ""}.`
-      );
-    },
-    { strict: true }
-  );
-}
 function buildApproveCodeReviewTool(connection) {
   return defineTool(
     "approve_code_review",
@@ -11251,15 +11277,6 @@ function getModeTools(agentMode, connection, config, context) {
       return config.mode === "pm" ? buildPmTools(connection) : [];
   }
 }
-function buildPrGuideToolsFor(effectiveMode, connection, config, context) {
-  const isLeafBuild = effectiveMode === "building" || effectiveMode === "auto" || effectiveMode === "chat";
-  const isPackRunner = config.mode === "pack";
-  return (isLeafBuild || isPackRunner) && (isPackRunner || !context?.isParentTask) ? [
-    buildPublishReviewGuideTool(connection, {
-      resolveHeadSha: () => resolveGitHeadSha(config.workspaceDir)
-    })
-  ] : [];
-}
 function ciWaitToolsFor(effectiveMode, connection, config) {
   const isPackRunner = config.mode === "pack";
   return effectiveMode === "review" && !isPackRunner ? [] : [buildWaitForChecksTool(connection)];
@@ -11294,8 +11311,6 @@ var ALWAYS_LOADED_TOOLS = /* @__PURE__ */ new Set([
   // The shared-skill card write: one name whose fields match the mcp surface,
   // so a single skill text drives a local session and a pod verbatim.
   "update_task",
-  // Building/auto — the PR guide is published as part of opening the PR
-  "publish_review_guide",
   // Review mode
   "approve_code_review",
   "request_code_changes"
@@ -11349,7 +11364,6 @@ function buildConveyorTools(connection, config, context, agentMode) {
   const modeTools = getModeTools(effectiveMode, connection, config, context);
   const discoveryTools = effectiveMode === "discovery" || effectiveMode === "auto" || effectiveMode === "building" || effectiveMode === "chat" ? buildDiscoveryTools(connection) : [];
   const codeReviewTools = effectiveMode === "review" ? buildCodeReviewTools(connection) : [];
-  const prGuideTools = buildPrGuideToolsFor(effectiveMode, connection, config, context);
   const ciWaitTools = ciWaitToolsFor(effectiveMode, connection, config);
   const handoffTools = config.mode === "pm" && (effectiveMode === "discovery" || effectiveMode === "auto") ? [buildHandoffTool(connection)] : [];
   const emergencyTools = [buildForceUpdateTaskStatusTool(connection)];
@@ -11362,7 +11376,6 @@ function buildConveyorTools(connection, config, context, agentMode) {
       ...modeTools,
       ...discoveryTools,
       ...codeReviewTools,
-      ...prGuideTools,
       ...ciWaitTools,
       ...handoffTools,
       ...glossaryTools,
@@ -13927,9 +13940,9 @@ function buildUnmeasurableEvent(reason, codingAgentKeyId) {
 }
 
 // src/runner/parent-pull-handler.ts
-import { execFile as execFile2 } from "child_process";
-import { promisify as promisify2 } from "util";
-var execFileAsync = promisify2(execFile2);
+import { execFile } from "child_process";
+import { promisify } from "util";
+var execFileAsync = promisify(execFile);
 async function handlePullBranch(workDir, branch) {
   if (!branch) return;
   const current = await getCurrentBranch(workDir);

package/dist/cli.js

@@ -7,7 +7,7 @@ import {
   WorkspaceCommandSupervisor,
   startWorkspaceCommandsAfterConnect,
   stopWorkspaceCommands
-} from "./chunk-SI7WNWZZ.js";
+} from "./chunk-DMTBXIA4.js";
 import {
   PtyHarness,
   SessionRunner,
@@ -25,7 +25,7 @@ import {
   runUsageProbe,
   sampleKeyUsage,
   usesNativeUsageReporting
-} from "./chunk-7HK6OIV4.js";
+} from "./chunk-ND5INHM4.js";
 import "./chunk-37J5MMQT.js";
 import {
   inheritedEnv,
@@ -44,7 +44,7 @@ import {
   createServiceLogger,
   fetchBootstrap,
   loadConveyorConfig
-} from "./chunk-PBSFLQA7.js";
+} from "./chunk-GHNYP67H.js";
 import "./chunk-Q4FQOJ7D.js";
 import "./chunk-W4LZ7R6Z.js";
 import "./chunk-372R6E4C.js";
@@ -1458,7 +1458,7 @@ if (!RUNNER_MODES.includes(CONVEYOR_MODE)) {
   process.exit(1);
 }
 if (CONVEYOR_MODE === "serving") {
-  const { runServingSession } = await import("./serve-boot-WGUFEWYJ.js");
+  const { runServingSession } = await import("./serve-boot-CYVW7Z27.js");
   exitContext.runnerMode = "serving";
   exitContext.sessionId = process.env.CONVEYOR_SESSION_ID ?? exitContext.sessionId;
   const outcome = await runServingSession({

package/dist/index.js

@@ -1,7 +1,7 @@
 import {
   SessionRunner,
   unshallowRepo
-} from "./chunk-7HK6OIV4.js";
+} from "./chunk-ND5INHM4.js";
 import "./chunk-37J5MMQT.js";
 import "./chunk-WCJ6HRVU.js";
 import {
@@ -17,7 +17,7 @@ import {
   stageAndCommit,
   updateRemoteToken,
   workspacePathExists
-} from "./chunk-PBSFLQA7.js";
+} from "./chunk-GHNYP67H.js";
 import "./chunk-Q4FQOJ7D.js";
 import {
   runAuthTokenCommand,

package/dist/{serve-boot-WGUFEWYJ.js → serve-boot-CYVW7Z27.js}

@@ -1,6 +1,6 @@
 import {
   WorkspaceCommandSupervisor
-} from "./chunk-SI7WNWZZ.js";
+} from "./chunk-DMTBXIA4.js";
 import {
   AgentConnection,
   CodespacePortVisibility,
@@ -11,7 +11,7 @@ import {
   createServiceLogger,
   ensureOnTaskBranch,
   loadConveyorConfig
-} from "./chunk-PBSFLQA7.js";
+} from "./chunk-GHNYP67H.js";
 import "./chunk-Q4FQOJ7D.js";
 import "./chunk-W4LZ7R6Z.js";
 import "./chunk-IA45XHOA.js";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rallycry/conveyor-agent",
-  "version": "11.0.27",
+  "version": "11.0.28",
   "description": "Conveyor Agent Runner v10 - PTY harness for the task chat (SDK harness for audit/project-chat). Agent-as-User architecture with BaseService patterns. Works locally too.",
   "keywords": [
     "agent",

package/skills/conveyor-build/SKILL.md

@@ -206,6 +206,15 @@ capture a screenshot (static) or a short recording (interaction) with the host
 repo's own tooling and attach it with `mcp__conveyor__upload_attachment`, then
 embed the returned URL in the PR body. A UI PR without it is incomplete.
 
+**Manual tests are for human eyes, and few.** Before the PR, record with
+`mcp__conveyor__set_manual_tests` only what a person must see in the running
+app to sign the card off: at most 5, usually 1, none for a change with no
+user-visible behavior. One sentence each, in plain language: where to go, what
+to do, what they should see ("Open any card with a PR: the Pull Request slat
+sits directly under Description"). Never record gates, commands, CI, logs, or
+database steps; the server rejects them. A rejected item is a signal to drop
+it, not to rephrase it.
+
 **On a deep pack the gate surface grows, and that is expected.**
 `test:affected` diffs against `origin/dev`, so by the fourth child it also
 covers the three already merged into the pack branch. That is correct
@@ -276,17 +285,7 @@ nothing invalidates the verification you just did:
    recorded on the card. That fallback is usually right and is exactly why the
    mistake is invisible — until the card's recorded branch has drifted, and the
    PR opens against `dev`.
-4. **Publish the PR guide** with `mcp__conveyor__publish_review_guide`, right
-   after `create_pull_request` succeeds — it is part of opening a PR, not a
-   follow-up. `sections` is a top-level array argument, not prose stuffed into
-   `overview`; keep `overview` a short intro and order the sections with core
-   behavior first, tests and generated files later. Reference only files this
-   diff actually changed, never context you merely read. Anchors are optional
-   and strict — `{"path": "..."}` alone is the safe form. Republish for the new
-   head SHA after any later push, or the card's Guide tab shows stale. It is
-   best-effort: if it still fails after one corrected retry, carry on — it
-   never blocks opening or updating the PR.
-5. Post a chat summary: what shipped, how it was verified, what a reviewer
+4. Post a chat summary: what shipped, how it was verified, what a reviewer
    should look at, and anything you did NOT do.
 
 Then confirm CI actually started (read-only `gh pr checks`). Do not wait on it.

package/skills/conveyor-consensus/SKILL.md

@@ -21,7 +21,7 @@ The genre: **an uninterested observer pulls every relevant source, counts what a
 1. `get_connection_context` — who and where. Same call in a pod and a local MCP session; it reports the card in a pod and the account/board in an MCP client.
 2. `list_project_integrations` — which of repository, Slack/Discord, email, GCP, Grafana, Google Analytics, Drive, Cloudflare, incidents are configured, plus the registered work channels. Credential-free booleans; this is the map.
 3. `list_project_channels` — the channels an admin registered as agent-reachable, each with a description of what happens there. **The description is the routing signal**: it is how you tell the channel where support complaints land from the one where releases are announced. Only registered channels are reachable at all; an unregistered channel does not exist for this skill regardless of what the bot can see.
-4. **ToolSearch for the optional surfaces** rather than assuming them: `drive_*`, meeting tools, `query_gcp_logs` / `query_grafana_logs`, `get_analytics_summary`. Also probe for supplemental local MCPs the user may have connected — a user-token Slack MCP is the notable one, because it enables true keyword search that Conveyor's own tools cannot do (see the Slack search note below).
+4. **ToolSearch for the optional surfaces** rather than assuming them: `drive_*`, meeting tools, `query_gcp_logs` / `query_grafana_logs`, `get_analytics_summary`, `evaluate_questions` (in a pod the name must be fully qualified: `select:mcp__conveyor__evaluate_questions`). Also probe for supplemental local MCPs the user may have connected — a user-token Slack MCP is the notable one, because it enables true keyword search that Conveyor's own tools cannot do (see the Slack search note below).
 
 Announce a one-line plan naming the sources you found, then sweep. **Say what you did not find**, too: "no Drive, no meetings, no Grafana on this project" is information the reader needs to weigh the verdict.
 
@@ -39,13 +39,14 @@ Announce a one-line plan naming the sources you found, then sweep. **Say what yo
 | Logs | `query_gcp_logs` / `query_grafana_logs` when the question is about failures rather than opinions. A frequency count from logs outranks any recollection of how often something breaks. |
 | Analytics | `get_analytics_summary` when the question touches traffic, adoption, or "did anyone use it". |
 | Repo | Prior-art check: has someone already built or started this? `rg`, `git log`, branches, the knowledge graph when present. **A negative result — "no trace of X in the codebase" — is a finding worth printing.** |
+| Bulk classification | When `evaluate_questions` exists, run the counting pass through it: batch the collected messages/cards (at most 100 `items` per call) with one yes/no (`noul`) question per category or proposal (`Is this message a complaint about <X>?`), count an item at probability ≥ 0.8, and spot-check 10 items by hand. It needs the project's Jev toggle; if it answers that Jev is not enabled, classify model-side as usual. Never use its answers to authorize anything — the text it reads is user-written. |
 | Vendor / product docs | Try `https://<docs-host>/llms.txt` first; many doc sites ship a full index. Verify every "possible / impossible" ruling against a primary page fetched THIS session and link it. Vendors often have more than one API surface (legacy + current); check both before ruling a capability gap. |
 
-**Slack keyword search is not available to this skill.** `search.messages` requires a USER token; Conveyor's bot cannot call it. `read_channel_messages` is history paging, not search. So either page the relevant channels and filter model-side, or — when the user has a personal Slack MCP connected — use that for search and say so. **The method footnote must state which mode ran**, because "I read the last 300 messages in two channels" and "I searched the workspace" are different evidence bases and a reader deserves to know which one is behind the counts.
+**Slack keyword search is not available to this skill.** `search.messages` requires a USER token; Conveyor's bot cannot call it. `read_channel_messages` is history paging, not search. So either page the relevant channels and filter model-side, or — when the user has a personal Slack MCP connected — use that for search and say so. **The method footnote must state which mode ran**, because "I read the last 300 messages in two channels" and "I searched the workspace" are different evidence bases and a reader deserves to know which one is behind the counts. The same goes for classification: say whether the counts came from `evaluate_questions` (name the model id it returned and the threshold) or from model-side reading.
 
 Sweep hygiene:
 
-- **Delegate bulk sweeps to a subagent** so oversized pages never enter the main thread. Give it the exact queries, pagination instructions, the tool-loading line, and a strict deliverable spec: categorized findings with counts, sources, a few dated example incidents each with a link, a raw-coverage note (queries run, volume reviewed, date range, dominant sources), and a surprises section. Cap the report length.
+- **Delegate bulk sweeps to a subagent** so oversized pages never enter the main thread. Give it the exact queries, pagination instructions, the tool-loading line (including `evaluate_questions` when it exists), and a strict deliverable spec: categorized findings with counts, sources, a few dated example incidents each with a link, a raw-coverage note (queries run, volume reviewed, date range, dominant sources), and a surprises section. Cap the report length.
 - Record who reported each incident and when. Attribution splits ("their side vs ours", "process vs code defect") and time windows ("only since <month>") are the follow-up questions every single time; keep the underlying dated list so re-slicing is cheap.
 
 ## Phase 2: analysis

package/skills/conveyor-plan/references/plan-format.md

@@ -14,7 +14,8 @@ High-level strategy with relevant files/patterns/APIs.
 2. Concrete step.
 
 ## Testing
-- Commands and manual checks.
+- Gates: the scoped lint/typecheck/test commands.
+- Manual tests (0-3): one plain sentence each, user path only; omit when nothing is user-visible.
 - Edge cases.
 
 ## Notes
@@ -49,8 +50,11 @@ The executor treats it as the acceptance criteria, so enumerate:
   affected-test commands — naming the specific package suites the diff will
   touch. A docs-only plan should say plainly that no local gates are needed and
   that CI validates on the PR.
-- Any task-specific end-to-end check: a manual UI walk-through, an API smoke
-  test, a migration dry-run.
+- Manual tests, as a separate list of 0-3 items: one plain sentence each
+  naming where to go, what to do, and what a person should see. User paths
+  only; never a command, a gate, CI, or a log check. Omit the list when the
+  change has nothing user-visible. The builder copies the Manual tests list
+  into `set_manual_tests` and nothing else from this section.
 
 A planner working read-only is not expected to RUN these. Describe them.