@integrity-labs/agt-cli 0.28.1002 → 0.28.1004

package/dist/bin/agt.js

@@ -40,12 +40,12 @@ import {
   success,
   table,
   warn
-} from "../chunk-5PJH4MHG.js";
+} from "../chunk-N4PG4TAD.js";
 import {
   getProjectDir,
   isSessionResumeDisabled,
   readDirectChatSessionState
-} from "../chunk-S6GIPDPI.js";
+} from "../chunk-HHBAQ6ZV.js";
 import {
   AnchorSessionClient,
   CHANNEL_REGISTRY,
@@ -79,7 +79,7 @@ import {
   serializeManifestForSlackCli,
   sessionFileExists,
   sessionTranscriptDir
-} from "../chunk-DGVA3EGS.js";
+} from "../chunk-532V7ZSC.js";
 import "../chunk-DHWNVVX4.js";
 import "../chunk-XWVM4KPK.js";
 import "../chunk-DGUM43GV.js";
@@ -5473,7 +5473,7 @@ import { execFileSync, execSync } from "child_process";
 import { existsSync as existsSync11, realpathSync as realpathSync2 } from "fs";
 import chalk18 from "chalk";
 import ora16 from "ora";
-var cliVersion = true ? "0.28.1002" : "dev";
+var cliVersion = true ? "0.28.1004" : "dev";
 async function fetchLatestVersion() {
   const host2 = getHost();
   if (!host2) return null;
@@ -6733,7 +6733,7 @@ function handleError(err) {
 }
 
 // src/bin/agt.ts
-var cliVersion2 = true ? "0.28.1002" : "dev";
+var cliVersion2 = true ? "0.28.1004" : "dev";
 var program = new Command();
 program.name("agt").description("Augmented CLI \u2014 agent provisioning and management").version(cliVersion2).option("--json", "Emit machine-readable JSON output (suppress spinners and colors)").option("--skip-update-check", "Skip the automatic update check on startup");
 program.hook("preAction", async (thisCommand, actionCommand) => {

package/dist/{chunk-DGVA3EGS.js → chunk-532V7ZSC.js}

@@ -3153,6 +3153,52 @@ var INTEGRATION_REGISTRY = [
       }]
     }
   },
+  {
+    // ENG-10854 / ADR-0110. The POLICY half of the catalog definition: this
+    // registry is what `isPremiumDefinition` reads, so the premium descriptor
+    // here is what makes the per-org opt-in required at all. The CUSTOMER-FACING
+    // half - the seeded `integration_definitions` row that the Add Integration
+    // picker lists - is deliberately NOT in this change. `/integrations/installs`
+    // reads the DB catalog with a status filter; this registry is not a listing.
+    //
+    // The order matters and is not bookkeeping: `isOrgEntitledToPremium` returns
+    // TRUE for a definition it does not know as premium, because a free
+    // integration has no opt-in concept. So a broker shipped ahead of this entry
+    // would serve an org that had never opted in. The broker additionally fails
+    // closed on `isPremiumDefinition` being false, so the two halves cannot come
+    // apart in that direction either.
+    id: "typed-evaluation",
+    name: "Typed Evaluation",
+    category: "knowledge",
+    description: "Ask closed questions about a piece of text and get calibrated answers back \u2014 yes/no, pick-one, or a position on your own scale \u2014 with a probability for each. It never writes text and cannot answer outside the set of options you give it, which is what makes it safe to point at content you do not trust. Typical uses: is this message a repeat request, which category does this belong to, how urgent does this sound. Answers arrive in about half a second, so an agent can branch on one without spending a model turn.",
+    // No customer credential exists: the call runs on Ninjafy's own gateway key.
+    supported_auth_types: ["none"],
+    capabilities: [
+      {
+        id: "typed-evaluation:evaluate",
+        name: "Evaluate Text",
+        description: "Ask closed questions about text the agent already holds and receive typed answers with probabilities (evaluate_text).",
+        access: "read"
+      }
+    ],
+    beta: true,
+    // ENG-8842's `included`, and the reason is the consent dialog rather than
+    // the price. `pricing` drives the org admin's acknowledgement modal, where
+    // `usage` renders as "manages and bills by usage" and `monthly` as a
+    // subscription. Both would be false statements here: a call costs about
+    // $0.000076 and nothing is metered or billed, so the opt-in exists to record
+    // a decision about a SUB-PROCESSOR, not about money.
+    //
+    // ADR-0110 Decision 11 keeps the door open deliberately: if a genuinely
+    // high-volume capability arrives later - bulk classification of a customer's
+    // own corpus - it is priced at THAT capability's grain and this definition
+    // stays `included`. Ninja Notes is the precedent (ENG-9278): it acquired a
+    // metered capability and did not flip.
+    premium: {
+      pricing: "included",
+      note: "Typed evaluation is included in your plan \u2014 nothing is metered per question. Turning it on is a decision about data, not cost: the text an agent asks about is sent to Typesafe, through our own gateway, to be scored. It is never used to train a model and we store only the answers, never the text."
+    }
+  },
   {
     // Ninja Notes (ADR-0063, ENG-8842). Meeting capture on the user's own Mac,
     // delivered to an agent as a tag-granted note the agent reads and summarises.
@@ -6273,6 +6319,33 @@ var FLAG_REGISTRY = [
     // "these hosts are unreadable". Both outlive this particular version.
     since: "0.28.667"
   },
+  {
+    // ADR-0110 Decision 5 (ENG-10854). Gates PLATFORM-INTERNAL consumers of the
+    // typed evaluation capability - inbound triage, kanban classification,
+    // memory-extraction sensitivity gating, and later the conversation
+    // evaluator. Nobody "installs" those; they are ours, so they are gated
+    // centrally.
+    //
+    // IT DOES NOT GATE THE AGENT TOOL, and that separation is the point. The
+    // agent tool's gate is the customer's own premium opt-in plus a per-agent
+    // install (Decision 4). Pointing this flag at the tool as well would give
+    // it two masters and let a staff flag silently override a decision the
+    // customer made about their own data reaching a sub-processor.
+    //
+    // Ships with no readers: the consumers land in later slices, and the flag
+    // exists now so the first of them has a gate to be born behind rather than
+    // acquiring one afterwards.
+    key: "typed-evaluation",
+    description: "ADR-0110: platform-internal consumers of the typed evaluation model (Jev, via our AI Gateway). OFF (default) = no internal consumer calls it; ON = consumers that have shipped may call, each still subject to the per-org quota, per-agent rate limit and audit row enforced in the broker. Does NOT gate the agent-facing tool, which is governed by the premium opt-in and a per-agent install instead. Customer orgs stay off until the Decision 9 vendor gates close (sub-processor listing with DPA coverage, the legal read on emotion inference, and a recorded disclosure).",
+    flagType: "boolean",
+    // Ships dark. Every consumer degrades to its current behaviour when off,
+    // so the safe value and the no-op value are the same one.
+    defaultValue: false,
+    // Turning it on starts sending customer text to an external sub-processor
+    // on paths the customer did not individually opt into, which is exactly the
+    // class ADR-0022 section 4 reserves the confirmation for.
+    sensitive: true
+  },
   {
     key: "conversation-eval-backend",
     description: "Backend for host-side conversation-success scoring AND memory extraction (ENG-6581), which share one scorer: anthropic-api = Haiku via a direct Anthropic Messages API fetch (no subprocess); claude-p = Haiku via the `claude -p` subprocess (reuses the agent auth, effectively free under Max but a heavy spawn); local = an OpenAI-compatible loopback endpoint (transcript never leaves the host). Auth for the anthropic-api path comes from AGT_CONV_EVAL_ANTHROPIC_API_KEY (falls back to ANTHROPIC_API_KEY); with no key that path fails closed (eval disabled) rather than reverting to claude-p. Enum.",
@@ -13459,6 +13532,100 @@ var SCHEDULED_TURN_MARKER_FILENAME = ".current-scheduled-turn.json";
 // ../../packages/core/dist/liveness/agent-liveness.js
 var FRESH_HEARTBEAT_THRESHOLD_MS = 2 * 60 * 1e3;
 
+// ../../packages/core/dist/claude-code-usage/host-usage-reading.js
+var HOST_USAGE_SOURCE = "cli_print_usage";
+function unknown(reason, readAt, cliVersion) {
+  return {
+    state: "unknown",
+    weekPct: null,
+    weekResetsAt: null,
+    sessionPct: null,
+    sessionResetsAt: null,
+    readAt,
+    cliVersion,
+    source: HOST_USAGE_SOURCE,
+    unknownReason: reason
+  };
+}
+function buildHostUsageReading(input) {
+  const { report, readAt, cliVersion, authMode, pollFailed } = input;
+  if (authMode != null && authMode !== "subscription") {
+    return {
+      state: "not_applicable",
+      weekPct: null,
+      weekResetsAt: null,
+      sessionPct: null,
+      sessionResetsAt: null,
+      readAt,
+      cliVersion,
+      source: HOST_USAGE_SOURCE,
+      unknownReason: null
+    };
+  }
+  if (pollFailed)
+    return unknown("poll_failed", readAt, cliVersion);
+  if (!report)
+    return unknown("unparseable_output", readAt, cliVersion);
+  if (!report.subscription) {
+    return {
+      state: "not_applicable",
+      weekPct: null,
+      weekResetsAt: null,
+      sessionPct: null,
+      sessionResetsAt: null,
+      readAt,
+      cliVersion,
+      source: HOST_USAGE_SOURCE,
+      unknownReason: null
+    };
+  }
+  if (!report.week && !report.session) {
+    return unknown("unparseable_output", readAt, cliVersion);
+  }
+  return {
+    state: "ok",
+    weekPct: report.week?.pct ?? null,
+    // ONLY a DATED reset is stored (CodeRabbit on #4988).
+    //
+    // Print mode has always rendered a date, so `hour_only` here means the CLI
+    // changed its render and `parseResetDateTime` reconstructed the day from an
+    // hour. That reconstruction can never be more than 24h out, and storing it
+    // in a column consumers treat as a measurement is precisely ENG-8472: every
+    // host whose render named the same hour landed on the same millisecond, and
+    // eleven days went into reading that as evidence of a shared subscription.
+    //
+    // The PERCENTAGE survives - it is a real reading either way, and dropping it
+    // would blind the fleet over a formatting change. Only the boundary is
+    // withheld, which degrades an alert to "cannot auto-close at reset" rather
+    // than to "asserts a reset instant nobody measured".
+    weekResetsAt: datedResetOnly(report.week),
+    sessionPct: report.session?.pct ?? null,
+    sessionResetsAt: datedResetOnly(report.session),
+    readAt,
+    cliVersion,
+    source: HOST_USAGE_SOURCE,
+    unknownReason: null
+  };
+}
+function datedResetOnly(window) {
+  if (!window || !window.resetsAt)
+    return null;
+  return window.resetPrecision === "dated" ? window.resetsAt : null;
+}
+var HOST_USAGE_POLL_INTERVAL_MS = 15 * 60 * 1e3;
+var READING_EXPIRY_INTERVALS = 2;
+var RESET_JITTER_TOLERANCE_MS = 5 * 60 * 1e3;
+function stabiliseResetInstant(previous, next, toleranceMs = RESET_JITTER_TOLERANCE_MS) {
+  if (!next)
+    return null;
+  if (!previous)
+    return next;
+  return Math.abs(next.getTime() - previous.getTime()) <= toleranceMs ? previous : next;
+}
+
+// ../../packages/core/dist/liveness/claude-usage-reading.js
+var STALE_READING_MS = HOST_USAGE_POLL_INTERVAL_MS * READING_EXPIRY_INTERVALS;
+
 // ../../packages/core/dist/claude-code-usage/reset-time.js
 var MONTHS = [
   "jan",
@@ -14057,96 +14224,6 @@ function parseUsageReport(text, now = /* @__PURE__ */ new Date()) {
   };
 }
 
-// ../../packages/core/dist/claude-code-usage/host-usage-reading.js
-var HOST_USAGE_SOURCE = "cli_print_usage";
-function unknown(reason, readAt, cliVersion) {
-  return {
-    state: "unknown",
-    weekPct: null,
-    weekResetsAt: null,
-    sessionPct: null,
-    sessionResetsAt: null,
-    readAt,
-    cliVersion,
-    source: HOST_USAGE_SOURCE,
-    unknownReason: reason
-  };
-}
-function buildHostUsageReading(input) {
-  const { report, readAt, cliVersion, authMode, pollFailed } = input;
-  if (authMode != null && authMode !== "subscription") {
-    return {
-      state: "not_applicable",
-      weekPct: null,
-      weekResetsAt: null,
-      sessionPct: null,
-      sessionResetsAt: null,
-      readAt,
-      cliVersion,
-      source: HOST_USAGE_SOURCE,
-      unknownReason: null
-    };
-  }
-  if (pollFailed)
-    return unknown("poll_failed", readAt, cliVersion);
-  if (!report)
-    return unknown("unparseable_output", readAt, cliVersion);
-  if (!report.subscription) {
-    return {
-      state: "not_applicable",
-      weekPct: null,
-      weekResetsAt: null,
-      sessionPct: null,
-      sessionResetsAt: null,
-      readAt,
-      cliVersion,
-      source: HOST_USAGE_SOURCE,
-      unknownReason: null
-    };
-  }
-  if (!report.week && !report.session) {
-    return unknown("unparseable_output", readAt, cliVersion);
-  }
-  return {
-    state: "ok",
-    weekPct: report.week?.pct ?? null,
-    // ONLY a DATED reset is stored (CodeRabbit on #4988).
-    //
-    // Print mode has always rendered a date, so `hour_only` here means the CLI
-    // changed its render and `parseResetDateTime` reconstructed the day from an
-    // hour. That reconstruction can never be more than 24h out, and storing it
-    // in a column consumers treat as a measurement is precisely ENG-8472: every
-    // host whose render named the same hour landed on the same millisecond, and
-    // eleven days went into reading that as evidence of a shared subscription.
-    //
-    // The PERCENTAGE survives - it is a real reading either way, and dropping it
-    // would blind the fleet over a formatting change. Only the boundary is
-    // withheld, which degrades an alert to "cannot auto-close at reset" rather
-    // than to "asserts a reset instant nobody measured".
-    weekResetsAt: datedResetOnly(report.week),
-    sessionPct: report.session?.pct ?? null,
-    sessionResetsAt: datedResetOnly(report.session),
-    readAt,
-    cliVersion,
-    source: HOST_USAGE_SOURCE,
-    unknownReason: null
-  };
-}
-function datedResetOnly(window) {
-  if (!window || !window.resetsAt)
-    return null;
-  return window.resetPrecision === "dated" ? window.resetsAt : null;
-}
-var HOST_USAGE_POLL_INTERVAL_MS = 15 * 60 * 1e3;
-var RESET_JITTER_TOLERANCE_MS = 5 * 60 * 1e3;
-function stabiliseResetInstant(previous, next, toleranceMs = RESET_JITTER_TOLERANCE_MS) {
-  if (!next)
-    return null;
-  if (!previous)
-    return next;
-  return Math.abs(next.getTime() - previous.getTime()) <= toleranceMs ? previous : next;
-}
-
 // ../../packages/core/dist/claude-code-usage/rate-limit-classifier.js
 var UNKNOWN_RATE_LIMIT = Object.freeze({
   verdict: "unknown",
@@ -15222,6 +15299,9 @@ export {
   isResolveError,
   deriveConsoleUrl,
   SCHEDULED_TURN_MARKER_FILENAME,
+  buildHostUsageReading,
+  HOST_USAGE_POLL_INTERVAL_MS,
+  stabiliseResetInstant,
   parseUsageBanner,
   bannerRemedyFor,
   formatRunMarker,
@@ -15231,9 +15311,6 @@ export {
   isEmptyTotals,
   attributeTranscriptUsageByRun,
   parseUsageReport,
-  buildHostUsageReading,
-  HOST_USAGE_POLL_INTERVAL_MS,
-  stabiliseResetInstant,
   UNKNOWN_RATE_LIMIT,
   pickNewerClassification,
   classifyTranscriptRateLimit,
@@ -15289,4 +15366,4 @@ export {
   peekCurrentSession,
   readDailySessionPin
 };
-//# sourceMappingURL=chunk-DGVA3EGS.js.map
+//# sourceMappingURL=chunk-532V7ZSC.js.map