@integrity-labs/agt-cli 0.27.150-test.15 → 0.27.150-test.17

package/dist/mcp/augmented-admin.js

@@ -21097,6 +21097,22 @@ var AdminDebugClient = class _AdminDebugClient {
   listAlerts(args) {
     return this.get("/admin/debug/alerts", _AdminDebugClient.cleanQuery(args));
   }
+  // ─────────────────── request-scoped customer grant (ENG-6279) ───────────────────
+  /**
+   * Open scoped READ access to a SELF-MANAGED customer org so the read tools can
+   * reach it (ENG-6279). POSTs to /admin/debug/grants, which mints the short-TTL,
+   * logged, customer-revocable read grant; the read tools then return that org's
+   * data for the grant's lifetime. Reuses `post()` — same token + 401-re-exchange
+   * flow as the reads. For an already-standing org (is_internal / fully_managed)
+   * the API returns `{ already_authorized: true }` instead of minting a redundant
+   * grant — not an error.
+   */
+  requestAccess(args) {
+    return this.post("/admin/debug/grants", {
+      target_org_id: args.target_org_id,
+      reason: args.reason
+    });
+  }
   // ─────────────────────── remedial actions (ENG-6197) ───────────────────────
   async post(path, body, retried = false) {
     const token = await this.getValidToken();
@@ -21167,6 +21183,12 @@ var listAlertsSchema = external_exports.object({
   open: external_exports.boolean().optional().describe("When true, only currently-open (unclosed) alerts."),
   limit: limitSchema
 });
+var requestAccessSchema = external_exports.object({
+  target_org_id: external_exports.string().min(1).max(64).describe("UUID of the customer organization to open scoped read access to."),
+  reason: external_exports.string().min(1).max(2e3).describe(
+    "Why access is needed \u2014 recorded on the customer-visible access log and shown to the org when they review or revoke it. Be specific."
+  )
+});
 var requestActionSchema = external_exports.object({
   action: external_exports.enum(["restart", "clear_pending_inbound"]).describe(
     "The remedial action. restart = bounce the agent (reversible; backlog replays). clear_pending_inbound = move stuck pending-inbound markers aside (drops queued inbound messages; they are NOT reprocessed)."
@@ -21181,7 +21203,7 @@ var checkActionStatusSchema = external_exports.object({
 // package.json
 var package_default = {
   name: "@integrity-labs/augmented-admin-mcp",
-  version: "0.1.0",
+  version: "0.1.1",
   description: "Augmented Admin Debug \u2014 Integrity Labs staff-only MCP server for cross-org agent/host/integration/alert diagnostics. Thin client over the Augmented API's /admin/debug/* surface; authority + the diagnostic projection live server-side. ENG-6195.",
   type: "module",
   private: true,
@@ -21249,7 +21271,7 @@ function formatError2(err) {
 var server = new McpServer({ name: "augmented-admin-mcp", version: package_default.version });
 server.tool(
   "debug_search_agents",
-  "Search managed end-user agents across all orgs for troubleshooting. Returns a diagnostic projection per agent (code_name, status, environment, risk_tier, org, heartbeat verdict) \u2014 never credentials or transcripts. Filter with { q, status, environment, limit }.",
+  "Search managed end-user agents across authorized orgs for troubleshooting. Returns a diagnostic projection per agent (code_name, status, environment, risk_tier, org, heartbeat verdict) \u2014 never credentials or transcripts. Authorized = IL-owned + fully-managed orgs (standing) plus any self-managed org you hold an active grant for (see debug_request_access). Filter with { q, status, environment, limit }.",
   searchAgentsSchema.shape,
   async (args) => {
     try {
@@ -21275,7 +21297,7 @@ server.tool(
 );
 server.tool(
   "debug_search_hosts",
-  "Search managed hosts across all orgs. Returns a projection per host (name, org, status, framework version, last-seen verdict, EC2 + Claude auth status) \u2014 no api-key fingerprints or secrets. Filter with { q, status, limit }.",
+  "Search managed hosts across authorized orgs. Returns a projection per host (name, org, status, framework version, last-seen verdict, EC2 + Claude auth status) \u2014 no api-key fingerprints or secrets. Filter with { q, status, limit }.",
   searchHostsSchema.shape,
   async (args) => {
     try {
@@ -21288,7 +21310,7 @@ server.tool(
 );
 server.tool(
   "debug_list_alerts",
-  "List recent platform alerts across all orgs (host-down, agent-stale, probe-timeout, auth-failed), including NULL-team host alerts. Returns a projection per alert (kind, severity, message, source, open/closed state). Filter with { severity, open, limit }.",
+  "List recent platform alerts across authorized orgs (host-down, agent-stale, probe-timeout, auth-failed), including NULL-team host alerts. Returns a projection per alert (kind, severity, message, source, open/closed state). Filter with { severity, open, limit }.",
   listAlertsSchema.shape,
   async (args) => {
     try {
@@ -21299,6 +21321,22 @@ server.tool(
     }
   }
 );
+server.tool(
+  "debug_request_access",
+  "Open scoped READ access to a SELF-MANAGED customer org so the read tools (debug_search_agents / debug_get_agent / debug_search_hosts / debug_list_alerts) can reach it. Use this when a read returns nothing for an org you have a legitimate support reason to troubleshoot. It mints a short-TTL, logged, customer-revocable read grant \u2014 the customer sees the access in their Support-access console and can revoke it, and it auto-expires; it does NOT grant any write/remedial capability. For an already-authorized org (Integrity-Labs-internal or fully-managed) this returns { already_authorized: true } instead of minting a redundant grant \u2014 just use the read tools. Pass { target_org_id, reason }; the reason is recorded verbatim on the customer-visible access log, so be specific.",
+  requestAccessSchema.shape,
+  async (args) => {
+    try {
+      const result = await client.requestAccess({
+        target_org_id: args.target_org_id,
+        reason: args.reason
+      });
+      return { content: [{ type: "text", text: JSON.stringify(result, null, 2) }] };
+    } catch (err) {
+      return { content: [{ type: "text", text: formatError2(err) }], isError: true };
+    }
+  }
+);
 server.tool(
   "request_action",
   `Request a HITL-gated remedial action on a customer agent: "restart" (bounce the agent \u2014 reversible, backlog replays) or "clear_pending_inbound" (move stuck pending-inbound markers aside \u2014 drops queued inbound messages, NOT reprocessed). A HUMAN must approve in Slack before anything happens, and only if writes are armed for this stage; in shadow mode the approval runs but nothing executes. Returns { request_id, status, write_mode, notification_status }. Surface a brief, jargon-free note to the user (e.g. "I've asked an admin to approve restarting that agent"). Pass { action, target_agent_id, reason }.`,

package/dist/{persistent-session-35PWSTLO.js → persistent-session-IKQLTZZ6.js}

@@ -25,9 +25,8 @@ import {
   takeAcpxExecFailureCount,
   takeZombieDetection,
   writePersistentClaudeWrapper
-} from "./chunk-7GKJZBTB.js";
-import "./chunk-WOOYOAPG.js";
-import "./chunk-354FAVQR.js";
+} from "./chunk-HXMLMIR4.js";
+import "./chunk-FZTGR2AQ.js";
 import "./chunk-XWVM4KPK.js";
 export {
   SEND_KEYS_ENTER_DELAY_MS,
@@ -57,4 +56,4 @@ export {
   takeZombieDetection,
   writePersistentClaudeWrapper
 };
-//# sourceMappingURL=persistent-session-35PWSTLO.js.map
+//# sourceMappingURL=persistent-session-IKQLTZZ6.js.map

package/dist/{responsiveness-probe-MA4M2QM4.js → responsiveness-probe-2K4QHOWW.js}

@@ -1,8 +1,7 @@
 import {
   paneLogPath
-} from "./chunk-7GKJZBTB.js";
-import "./chunk-WOOYOAPG.js";
-import "./chunk-354FAVQR.js";
+} from "./chunk-HXMLMIR4.js";
+import "./chunk-FZTGR2AQ.js";
 import "./chunk-XWVM4KPK.js";
 
 // src/lib/responsiveness-probe.ts
@@ -155,4 +154,4 @@ export {
   livePendingInboundOldestAgeSeconds,
   oldestLivePendingInboundMtimeMs
 };
-//# sourceMappingURL=responsiveness-probe-MA4M2QM4.js.map
+//# sourceMappingURL=responsiveness-probe-2K4QHOWW.js.map

package/dist/{responsiveness-probe-MA4M2QM4.js.map → responsiveness-probe-2K4QHOWW.js.map}

@@ -1 +1 @@
-{"version":3,"sources":["../src/lib/responsiveness-probe.ts"],"sourcesContent":["/**\n * ENG-5399 — Tier 1 responsiveness probe (manager-side).\n *\n * Cheap, fast-cadence canary that catches \"agent went silent\" inside\n * minutes, well before the existing synthetic-probe cron's ~35 min\n * staleness window (`SyntheticReplyAgeSeconds`, ENG-5122).\n *\n * Mechanism: for each managed agent, read the mtime of the agent's\n * `pane.log` and report `now - mtime` as `PaneActivityAgeSeconds` via\n * a new `/host/responsiveness-probe` endpoint. `pane.log` is the\n * tmux pipe-pane sink set up by `setupPaneLog()` — any visible\n * activity (assistant turns, tool calls, in-place progress\n * heartbeats) bumps its mtime. A silent agent has a steadily\n * climbing age that lands in CloudWatch and trips a per-agent alarm.\n *\n * ENG-6017 adds a second per-agent signal on the same cadence:\n * `pending_inbound_oldest_age_seconds` — the age of the oldest marker\n * file across the agent's `*-pending-inbound/` directories (written by\n * the channel MCP servers for inbounds awaiting delivery). This is the\n * one artifact of the \"message typed but never submitted\" failure mode\n * that every other canary is blind to: in the koda incident\n * (2026-06-04) an operator Slack DM sat undelivered for 40+ minutes\n * while pane-activity stayed fresh (health checks), synthetic probes\n * were answered by the one-shot fallback, and heartbeat/session-alive\n * only reflect manager health. The field is OMITTED (not zero) when the\n * agent has no pending-inbound markers — the API treats absent as\n * \"no signal\", never as \"healthy\" (absent-vs-zero matters for\n * mixed-version fleets where old CLIs don't report it at all).\n *\n * Run from `pollCycle()` in `manager-worker.ts` on a configurable\n * interval (default 5 min via `AUGMENTED_RESPONSIVENESS_INTERVAL_MS`).\n */\n\nimport { mkdirSync, readdirSync, readFileSync, renameSync, statSync } from 'node:fs';\nimport { dirname, join } from 'node:path';\nimport { paneLogPath } from './persistent-session.js';\n\nexport interface ResponsivenessProbeResult {\n  code_name: string;\n  pane_activity_age_seconds: number;\n  /**\n   * ENG-6017: age (s) of the oldest marker file across the agent's\n   * `*-pending-inbound/` directories. Omitted when no markers exist —\n   * absent means \"no signal\", NOT \"zero / healthy\".\n   */\n  pending_inbound_oldest_age_seconds?: number;\n}\n\nconst DEFAULT_INTERVAL_MS = 5 * 60 * 1000;\n\nexport function getResponsivenessIntervalMs(): number {\n  const raw = process.env.AUGMENTED_RESPONSIVENESS_INTERVAL_MS;\n  if (!raw) return DEFAULT_INTERVAL_MS;\n  const parsed = Number.parseInt(raw, 10);\n  return Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_INTERVAL_MS;\n}\n\n/**\n * ENG-6017: oldest pending-inbound marker mtime (ms epoch) for an agent,\n * or null when the agent has no markers / no pending-inbound dirs.\n *\n * The channel MCP servers (slack-channel, telegram-channel, …) write one\n * marker file per inbound into `~/.augmented/<codeName>/<channel>-pending-\n * inbound/` and clear it when the agent acknowledges the message. The\n * directory layout is the contract here — read-only, no IPC with the MCP\n * (the MCP and CLI release independently; file mtimes need no protocol).\n *\n * ENG-6072: only plain, non-hidden files count as markers. The msteams MCP\n * keeps `.markers/` and `.processed/` housekeeping SUBDIRECTORIES inside its\n * pending-inbound dir; their mtimes never advance, so statting every dirent\n * made the gauge climb forever and fired pending-inbound-stale on agents with\n * zero stranded messages (kylie ~3.4d / scout ~34h false ALARMs the moment\n * ENG-6023 activated the alarm). Dot-entries are skipped wholesale — the\n * hidden namespace is reserved for MCP bookkeeping, never for markers.\n */\nfunction oldestPendingInboundMtimeMs(agentHomeDir: string): number | null {\n  let oldest: number | null = null;\n  let entries;\n  try {\n    entries = readdirSync(agentHomeDir, { withFileTypes: true });\n  } catch {\n    return null; // agent home missing — nothing to report\n  }\n  for (const entry of entries) {\n    if (!entry.isDirectory() || !entry.name.endsWith('-pending-inbound')) continue;\n    const dir = join(agentHomeDir, entry.name);\n    let files;\n    try {\n      files = readdirSync(dir, { withFileTypes: true });\n    } catch {\n      continue;\n    }\n    for (const file of files) {\n      if (!file.isFile() || file.name.startsWith('.')) continue;\n      try {\n        const mtimeMs = statSync(join(dir, file.name)).mtimeMs;\n        if (oldest === null || mtimeMs < oldest) oldest = mtimeMs;\n      } catch {\n        // Marker drained between readdir and stat — that's the happy path.\n      }\n    }\n  }\n  return oldest;\n}\n\n/**\n * ENG-6160: classify a marker file for the LIVE-inbound scan.\n *  - `true`  → flagged `\"undeliverable\": true` (dead-letter, exclude).\n *  - `null`  → vanished mid-scan (ENOENT) — drained between stat and read, the\n *              happy path; exclude it rather than count an already-gone file.\n *  - `false` → still present but malformed / unreadable for another reason —\n *              treated as LIVE so a corrupt marker can never mask a real wedge.\n */\nfunction isUndeliverableMarker(markerPath: string): boolean | null {\n  try {\n    const parsed = JSON.parse(readFileSync(markerPath, 'utf8')) as { undeliverable?: unknown };\n    return parsed?.undeliverable === true;\n  } catch (error) {\n    return (error as NodeJS.ErrnoException).code === 'ENOENT' ? null : false;\n  }\n}\n\n/**\n * ENG-6160: oldest *LIVE* pending-inbound marker mtime (ms epoch) for an agent,\n * or null when there is no live marker. \"Live\" excludes:\n *\n *  - markers older than `sessionStartMs` — a marker written before the current\n *    session started is a leftover from a PREVIOUS session and cannot mean\n *    *this* session is failing to drain. This is the load-bearing exclusion:\n *    without it, an orphan marker survives a fresh respawn and the wedge\n *    detector re-fires forever on a healthy idle agent (the sherlock enforce\n *    loop, 2026-06-08: `inboundAge=3389s` on a `● Ready.` session).\n *  - markers flagged `undeliverable: true` — already dead-lettered by the channel.\n *\n * Distinct from `oldestPendingInboundMtimeMs` (which counts ALL markers and\n * feeds the ENG-6017 `pending-inbound-stale` CloudWatch alarm — that alarm\n * *wants* to fire on a stuck inbound, so its semantics must NOT change). This\n * variant is wedge-detection-only.\n */\nexport function oldestLivePendingInboundMtimeMs(\n  agentHomeDir: string,\n  opts: { sessionStartMs?: number | null } = {},\n): number | null {\n  const sessionStartMs = opts.sessionStartMs ?? null;\n  let oldest: number | null = null;\n  let entries;\n  try {\n    entries = readdirSync(agentHomeDir, { withFileTypes: true });\n  } catch {\n    return null;\n  }\n  for (const entry of entries) {\n    if (!entry.isDirectory() || !entry.name.endsWith('-pending-inbound')) continue;\n    const dir = join(agentHomeDir, entry.name);\n    let files;\n    try {\n      files = readdirSync(dir, { withFileTypes: true });\n    } catch {\n      continue;\n    }\n    for (const file of files) {\n      if (!file.isFile() || file.name.startsWith('.')) continue;\n      const full = join(dir, file.name);\n      let mtimeMs: number;\n      try {\n        mtimeMs = statSync(full).mtimeMs;\n      } catch {\n        continue; // drained between readdir and stat — happy path\n      }\n      if (sessionStartMs !== null && mtimeMs < sessionStartMs) continue; // pre-session leftover\n      const undeliverable = isUndeliverableMarker(full);\n      if (undeliverable === null) continue; // vanished between stat and read — drained, exclude\n      if (undeliverable) continue; // already dead-lettered\n      if (oldest === null || mtimeMs < oldest) oldest = mtimeMs;\n    }\n  }\n  return oldest;\n}\n\n/**\n * ENG-6160: age (s) of the oldest LIVE pending-inbound marker for an agent, or\n * null when none. The wedge detector uses this instead of the alarm-facing\n * `pending_inbound_oldest_age_seconds` so a stale/dead-letter marker can't\n * false-fire a respawn.\n */\nexport function livePendingInboundOldestAgeSeconds(\n  codeName: string,\n  sessionStartMs: number | null,\n  now: Date = new Date(),\n): number | null {\n  const oldest = oldestLivePendingInboundMtimeMs(dirname(paneLogPath(codeName)), { sessionStartMs });\n  if (oldest === null) return null;\n  return Math.max(0, Math.floor((now.getTime() - oldest) / 1000));\n}\n\n/**\n * ENG-6160: move every pending-inbound marker for an agent aside into a sibling\n * `<channel>-pending-inbound-stale/` directory (NOT silently deleted — the\n * payload pointer is preserved for forensics), returning the count moved.\n *\n * Called on a force-fresh wedge respawn: the markers belonged to the wedged\n * session that is being torn down; the fresh session cannot meaningfully\n * process a stale, out-of-context message, and leaving them on disk both keeps\n * the ENG-6017 alarm lit and (pre-ENG-6160) re-fed the wedge loop. The stale\n * dir does not end in `-pending-inbound`, so neither the probe nor this scan\n * re-counts moved markers.\n */\nexport function deadLetterPendingInbound(codeName: string, _now: Date = new Date()): number {\n  const home = dirname(paneLogPath(codeName));\n  let moved = 0;\n  let entries;\n  try {\n    entries = readdirSync(home, { withFileTypes: true });\n  } catch {\n    return 0;\n  }\n  for (const entry of entries) {\n    if (!entry.isDirectory() || !entry.name.endsWith('-pending-inbound')) continue;\n    const dir = join(home, entry.name);\n    const deadDir = join(home, `${entry.name}-stale`);\n    let files;\n    try {\n      files = readdirSync(dir, { withFileTypes: true });\n    } catch {\n      continue;\n    }\n    for (const file of files) {\n      if (!file.isFile() || file.name.startsWith('.')) continue;\n      try {\n        mkdirSync(deadDir, { recursive: true });\n        renameSync(join(dir, file.name), join(deadDir, file.name));\n        moved++;\n      } catch {\n        // best-effort — a marker that vanished or can't move is left as-is\n      }\n    }\n  }\n  return moved;\n}\n\n/**\n * Compute the pane.log age for each agent. Missing or unreadable\n * pane.log returns null — the caller should drop those entries\n * rather than fabricate a \"fresh\" or \"ancient\" value. A missing\n * file means the agent has never spawned in this manager generation,\n * which is a separate problem covered by SessionAliveAgeSeconds.\n */\nexport function collectResponsivenessProbes(\n  codeNames: string[],\n  now: Date = new Date(),\n): ResponsivenessProbeResult[] {\n  const nowMs = now.getTime();\n  const results: ResponsivenessProbeResult[] = [];\n  for (const codeName of codeNames) {\n    try {\n      const panePath = paneLogPath(codeName);\n      const mtimeMs = statSync(panePath).mtimeMs;\n      const ageSeconds = Math.max(0, Math.floor((nowMs - mtimeMs) / 1000));\n      const result: ResponsivenessProbeResult = {\n        code_name: codeName,\n        pane_activity_age_seconds: ageSeconds,\n      };\n      // ENG-6017: piggyback the pending-inbound drain-age scan on the same\n      // cadence. Field omitted (not 0) when there are no markers.\n      const oldestMarkerMs = oldestPendingInboundMtimeMs(dirname(panePath));\n      if (oldestMarkerMs !== null) {\n        result.pending_inbound_oldest_age_seconds = Math.max(\n          0,\n          Math.floor((nowMs - oldestMarkerMs) / 1000),\n        );\n      }\n      results.push(result);\n    } catch {\n      // No pane.log yet (fresh agent, never spawned) — skip. The\n      // session-alive monitor already covers the \"should be running\n      // but isn't\" case.\n    }\n  }\n  return results;\n}\n"],"mappings":";;;;;;;;AAiCA,SAAS,WAAW,aAAa,cAAc,YAAY,gBAAgB;AAC3E,SAAS,SAAS,YAAY;AAc9B,IAAM,sBAAsB,IAAI,KAAK;AAE9B,SAAS,8BAAsC;AACpD,QAAM,MAAM,QAAQ,IAAI;AACxB,MAAI,CAAC,IAAK,QAAO;AACjB,QAAM,SAAS,OAAO,SAAS,KAAK,EAAE;AACtC,SAAO,OAAO,SAAS,MAAM,KAAK,SAAS,IAAI,SAAS;AAC1D;AAoBA,SAAS,4BAA4B,cAAqC;AACxE,MAAI,SAAwB;AAC5B,MAAI;AACJ,MAAI;AACF,cAAU,YAAY,cAAc,EAAE,eAAe,KAAK,CAAC;AAAA,EAC7D,QAAQ;AACN,WAAO;AAAA,EACT;AACA,aAAW,SAAS,SAAS;AAC3B,QAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,KAAK,SAAS,kBAAkB,EAAG;AACtE,UAAM,MAAM,KAAK,cAAc,MAAM,IAAI;AACzC,QAAI;AACJ,QAAI;AACF,cAAQ,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;AAAA,IAClD,QAAQ;AACN;AAAA,IACF;AACA,eAAW,QAAQ,OAAO;AACxB,UAAI,CAAC,KAAK,OAAO,KAAK,KAAK,KAAK,WAAW,GAAG,EAAG;AACjD,UAAI;AACF,cAAM,UAAU,SAAS,KAAK,KAAK,KAAK,IAAI,CAAC,EAAE;AAC/C,YAAI,WAAW,QAAQ,UAAU,OAAQ,UAAS;AAAA,MACpD,QAAQ;AAAA,MAER;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAUA,SAAS,sBAAsB,YAAoC;AACjE,MAAI;AACF,UAAM,SAAS,KAAK,MAAM,aAAa,YAAY,MAAM,CAAC;AAC1D,WAAO,QAAQ,kBAAkB;AAAA,EACnC,SAAS,OAAO;AACd,WAAQ,MAAgC,SAAS,WAAW,OAAO;AAAA,EACrE;AACF;AAmBO,SAAS,gCACd,cACA,OAA2C,CAAC,GAC7B;AACf,QAAM,iBAAiB,KAAK,kBAAkB;AAC9C,MAAI,SAAwB;AAC5B,MAAI;AACJ,MAAI;AACF,cAAU,YAAY,cAAc,EAAE,eAAe,KAAK,CAAC;AAAA,EAC7D,QAAQ;AACN,WAAO;AAAA,EACT;AACA,aAAW,SAAS,SAAS;AAC3B,QAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,KAAK,SAAS,kBAAkB,EAAG;AACtE,UAAM,MAAM,KAAK,cAAc,MAAM,IAAI;AACzC,QAAI;AACJ,QAAI;AACF,cAAQ,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;AAAA,IAClD,QAAQ;AACN;AAAA,IACF;AACA,eAAW,QAAQ,OAAO;AACxB,UAAI,CAAC,KAAK,OAAO,KAAK,KAAK,KAAK,WAAW,GAAG,EAAG;AACjD,YAAM,OAAO,KAAK,KAAK,KAAK,IAAI;AAChC,UAAI;AACJ,UAAI;AACF,kBAAU,SAAS,IAAI,EAAE;AAAA,MAC3B,QAAQ;AACN;AAAA,MACF;AACA,UAAI,mBAAmB,QAAQ,UAAU,eAAgB;AACzD,YAAM,gBAAgB,sBAAsB,IAAI;AAChD,UAAI,kBAAkB,KAAM;AAC5B,UAAI,cAAe;AACnB,UAAI,WAAW,QAAQ,UAAU,OAAQ,UAAS;AAAA,IACpD;AAAA,EACF;AACA,SAAO;AACT;AAQO,SAAS,mCACd,UACA,gBACA,MAAY,oBAAI,KAAK,GACN;AACf,QAAM,SAAS,gCAAgC,QAAQ,YAAY,QAAQ,CAAC,GAAG,EAAE,eAAe,CAAC;AACjG,MAAI,WAAW,KAAM,QAAO;AAC5B,SAAO,KAAK,IAAI,GAAG,KAAK,OAAO,IAAI,QAAQ,IAAI,UAAU,GAAI,CAAC;AAChE;AAcO,SAAS,yBAAyB,UAAkB,OAAa,oBAAI,KAAK,GAAW;AAC1F,QAAM,OAAO,QAAQ,YAAY,QAAQ,CAAC;AAC1C,MAAI,QAAQ;AACZ,MAAI;AACJ,MAAI;AACF,cAAU,YAAY,MAAM,EAAE,eAAe,KAAK,CAAC;AAAA,EACrD,QAAQ;AACN,WAAO;AAAA,EACT;AACA,aAAW,SAAS,SAAS;AAC3B,QAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,KAAK,SAAS,kBAAkB,EAAG;AACtE,UAAM,MAAM,KAAK,MAAM,MAAM,IAAI;AACjC,UAAM,UAAU,KAAK,MAAM,GAAG,MAAM,IAAI,QAAQ;AAChD,QAAI;AACJ,QAAI;AACF,cAAQ,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;AAAA,IAClD,QAAQ;AACN;AAAA,IACF;AACA,eAAW,QAAQ,OAAO;AACxB,UAAI,CAAC,KAAK,OAAO,KAAK,KAAK,KAAK,WAAW,GAAG,EAAG;AACjD,UAAI;AACF,kBAAU,SAAS,EAAE,WAAW,KAAK,CAAC;AACtC,mBAAW,KAAK,KAAK,KAAK,IAAI,GAAG,KAAK,SAAS,KAAK,IAAI,CAAC;AACzD;AAAA,MACF,QAAQ;AAAA,MAER;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AASO,SAAS,4BACd,WACA,MAAY,oBAAI,KAAK,GACQ;AAC7B,QAAM,QAAQ,IAAI,QAAQ;AAC1B,QAAM,UAAuC,CAAC;AAC9C,aAAW,YAAY,WAAW;AAChC,QAAI;AACF,YAAM,WAAW,YAAY,QAAQ;AACrC,YAAM,UAAU,SAAS,QAAQ,EAAE;AACnC,YAAM,aAAa,KAAK,IAAI,GAAG,KAAK,OAAO,QAAQ,WAAW,GAAI,CAAC;AACnE,YAAM,SAAoC;AAAA,QACxC,WAAW;AAAA,QACX,2BAA2B;AAAA,MAC7B;AAGA,YAAM,iBAAiB,4BAA4B,QAAQ,QAAQ,CAAC;AACpE,UAAI,mBAAmB,MAAM;AAC3B,eAAO,qCAAqC,KAAK;AAAA,UAC/C;AAAA,UACA,KAAK,OAAO,QAAQ,kBAAkB,GAAI;AAAA,QAC5C;AAAA,MACF;AACA,cAAQ,KAAK,MAAM;AAAA,IACrB,QAAQ;AAAA,IAIR;AAAA,EACF;AACA,SAAO;AACT;","names":[]}
+{"version":3,"sources":["../src/lib/responsiveness-probe.ts"],"sourcesContent":["/**\n * ENG-5399 — Tier 1 responsiveness probe (manager-side).\n *\n * Cheap, fast-cadence canary that catches \"agent went silent\" inside\n * minutes, well before the existing synthetic-probe cron's ~35 min\n * staleness window (`SyntheticReplyAgeSeconds`, ENG-5122).\n *\n * Mechanism: for each managed agent, read the mtime of the agent's\n * `pane.log` and report `now - mtime` as `PaneActivityAgeSeconds` via\n * a new `/host/responsiveness-probe` endpoint. `pane.log` is the\n * tmux pipe-pane sink set up by `setupPaneLog()` — any visible\n * activity (assistant turns, tool calls, in-place progress\n * heartbeats) bumps its mtime. A silent agent has a steadily\n * climbing age that lands in CloudWatch and trips a per-agent alarm.\n *\n * ENG-6017 adds a second per-agent signal on the same cadence:\n * `pending_inbound_oldest_age_seconds` — the age of the oldest marker\n * file across the agent's `*-pending-inbound/` directories (written by\n * the channel MCP servers for inbounds awaiting delivery). This is the\n * one artifact of the \"message typed but never submitted\" failure mode\n * that every other canary is blind to: in the koda incident\n * (2026-06-04) an operator Slack DM sat undelivered for 40+ minutes\n * while pane-activity stayed fresh (health checks), synthetic probes\n * were answered by the one-shot fallback, and heartbeat/session-alive\n * only reflect manager health. The field is OMITTED (not zero) when the\n * agent has no pending-inbound markers — the API treats absent as\n * \"no signal\", never as \"healthy\" (absent-vs-zero matters for\n * mixed-version fleets where old CLIs don't report it at all).\n *\n * Run from `pollCycle()` in `manager-worker.ts` on a configurable\n * interval (default 5 min via `AUGMENTED_RESPONSIVENESS_INTERVAL_MS`).\n */\n\nimport { mkdirSync, readdirSync, readFileSync, renameSync, statSync } from 'node:fs';\nimport { dirname, join } from 'node:path';\nimport { paneLogPath } from './persistent-session.js';\n\nexport interface ResponsivenessProbeResult {\n  code_name: string;\n  pane_activity_age_seconds: number;\n  /**\n   * ENG-6017: age (s) of the oldest marker file across the agent's\n   * `*-pending-inbound/` directories. Omitted when no markers exist —\n   * absent means \"no signal\", NOT \"zero / healthy\".\n   */\n  pending_inbound_oldest_age_seconds?: number;\n}\n\nconst DEFAULT_INTERVAL_MS = 5 * 60 * 1000;\n\nexport function getResponsivenessIntervalMs(): number {\n  const raw = process.env.AUGMENTED_RESPONSIVENESS_INTERVAL_MS;\n  if (!raw) return DEFAULT_INTERVAL_MS;\n  const parsed = Number.parseInt(raw, 10);\n  return Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_INTERVAL_MS;\n}\n\n/**\n * ENG-6017: oldest pending-inbound marker mtime (ms epoch) for an agent,\n * or null when the agent has no markers / no pending-inbound dirs.\n *\n * The channel MCP servers (slack-channel, telegram-channel, …) write one\n * marker file per inbound into `~/.augmented/<codeName>/<channel>-pending-\n * inbound/` and clear it when the agent acknowledges the message. The\n * directory layout is the contract here — read-only, no IPC with the MCP\n * (the MCP and CLI release independently; file mtimes need no protocol).\n *\n * ENG-6072: only plain, non-hidden files count as markers. The msteams MCP\n * keeps `.markers/` and `.processed/` housekeeping SUBDIRECTORIES inside its\n * pending-inbound dir; their mtimes never advance, so statting every dirent\n * made the gauge climb forever and fired pending-inbound-stale on agents with\n * zero stranded messages (kylie ~3.4d / scout ~34h false ALARMs the moment\n * ENG-6023 activated the alarm). Dot-entries are skipped wholesale — the\n * hidden namespace is reserved for MCP bookkeeping, never for markers.\n */\nfunction oldestPendingInboundMtimeMs(agentHomeDir: string): number | null {\n  let oldest: number | null = null;\n  let entries;\n  try {\n    entries = readdirSync(agentHomeDir, { withFileTypes: true });\n  } catch {\n    return null; // agent home missing — nothing to report\n  }\n  for (const entry of entries) {\n    if (!entry.isDirectory() || !entry.name.endsWith('-pending-inbound')) continue;\n    const dir = join(agentHomeDir, entry.name);\n    let files;\n    try {\n      files = readdirSync(dir, { withFileTypes: true });\n    } catch {\n      continue;\n    }\n    for (const file of files) {\n      if (!file.isFile() || file.name.startsWith('.')) continue;\n      try {\n        const mtimeMs = statSync(join(dir, file.name)).mtimeMs;\n        if (oldest === null || mtimeMs < oldest) oldest = mtimeMs;\n      } catch {\n        // Marker drained between readdir and stat — that's the happy path.\n      }\n    }\n  }\n  return oldest;\n}\n\n/**\n * ENG-6160: classify a marker file for the LIVE-inbound scan.\n *  - `true`  → flagged `\"undeliverable\": true` (dead-letter, exclude).\n *  - `null`  → vanished mid-scan (ENOENT) — drained between stat and read, the\n *              happy path; exclude it rather than count an already-gone file.\n *  - `false` → still present but malformed / unreadable for another reason —\n *              treated as LIVE so a corrupt marker can never mask a real wedge.\n */\nfunction isUndeliverableMarker(markerPath: string): boolean | null {\n  try {\n    const parsed = JSON.parse(readFileSync(markerPath, 'utf8')) as { undeliverable?: unknown };\n    return parsed?.undeliverable === true;\n  } catch (error) {\n    return (error as NodeJS.ErrnoException).code === 'ENOENT' ? null : false;\n  }\n}\n\n/**\n * ENG-6160: oldest *LIVE* pending-inbound marker mtime (ms epoch) for an agent,\n * or null when there is no live marker. \"Live\" excludes:\n *\n *  - markers older than `sessionStartMs` — a marker written before the current\n *    session started is a leftover from a PREVIOUS session and cannot mean\n *    *this* session is failing to drain. This is the load-bearing exclusion:\n *    without it, an orphan marker survives a fresh respawn and the wedge\n *    detector re-fires forever on a healthy idle agent (the sherlock enforce\n *    loop, 2026-06-08: `inboundAge=3389s` on a `● Ready.` session).\n *  - markers flagged `undeliverable: true` — already dead-lettered by the channel.\n *\n * Distinct from `oldestPendingInboundMtimeMs` (which counts ALL markers and\n * feeds the ENG-6017 `pending-inbound-stale` CloudWatch alarm — that alarm\n * *wants* to fire on a stuck inbound, so its semantics must NOT change). This\n * variant is wedge-detection-only.\n */\nexport function oldestLivePendingInboundMtimeMs(\n  agentHomeDir: string,\n  opts: { sessionStartMs?: number | null } = {},\n): number | null {\n  const sessionStartMs = opts.sessionStartMs ?? null;\n  let oldest: number | null = null;\n  let entries;\n  try {\n    entries = readdirSync(agentHomeDir, { withFileTypes: true });\n  } catch {\n    return null;\n  }\n  for (const entry of entries) {\n    if (!entry.isDirectory() || !entry.name.endsWith('-pending-inbound')) continue;\n    const dir = join(agentHomeDir, entry.name);\n    let files;\n    try {\n      files = readdirSync(dir, { withFileTypes: true });\n    } catch {\n      continue;\n    }\n    for (const file of files) {\n      if (!file.isFile() || file.name.startsWith('.')) continue;\n      const full = join(dir, file.name);\n      let mtimeMs: number;\n      try {\n        mtimeMs = statSync(full).mtimeMs;\n      } catch {\n        continue; // drained between readdir and stat — happy path\n      }\n      if (sessionStartMs !== null && mtimeMs < sessionStartMs) continue; // pre-session leftover\n      const undeliverable = isUndeliverableMarker(full);\n      if (undeliverable === null) continue; // vanished between stat and read — drained, exclude\n      if (undeliverable) continue; // already dead-lettered\n      if (oldest === null || mtimeMs < oldest) oldest = mtimeMs;\n    }\n  }\n  return oldest;\n}\n\n/**\n * ENG-6160: age (s) of the oldest LIVE pending-inbound marker for an agent, or\n * null when none. The wedge detector uses this instead of the alarm-facing\n * `pending_inbound_oldest_age_seconds` so a stale/dead-letter marker can't\n * false-fire a respawn.\n */\nexport function livePendingInboundOldestAgeSeconds(\n  codeName: string,\n  sessionStartMs: number | null,\n  now: Date = new Date(),\n): number | null {\n  const oldest = oldestLivePendingInboundMtimeMs(dirname(paneLogPath(codeName)), { sessionStartMs });\n  if (oldest === null) return null;\n  return Math.max(0, Math.floor((now.getTime() - oldest) / 1000));\n}\n\n/**\n * ENG-6160: move every pending-inbound marker for an agent aside into a sibling\n * `<channel>-pending-inbound-stale/` directory (NOT silently deleted — the\n * payload pointer is preserved for forensics), returning the count moved.\n *\n * Called on a force-fresh wedge respawn: the markers belonged to the wedged\n * session that is being torn down; the fresh session cannot meaningfully\n * process a stale, out-of-context message, and leaving them on disk both keeps\n * the ENG-6017 alarm lit and (pre-ENG-6160) re-fed the wedge loop. The stale\n * dir does not end in `-pending-inbound`, so neither the probe nor this scan\n * re-counts moved markers.\n */\nexport function deadLetterPendingInbound(codeName: string, _now: Date = new Date()): number {\n  const home = dirname(paneLogPath(codeName));\n  let moved = 0;\n  let entries;\n  try {\n    entries = readdirSync(home, { withFileTypes: true });\n  } catch {\n    return 0;\n  }\n  for (const entry of entries) {\n    if (!entry.isDirectory() || !entry.name.endsWith('-pending-inbound')) continue;\n    const dir = join(home, entry.name);\n    const deadDir = join(home, `${entry.name}-stale`);\n    let files;\n    try {\n      files = readdirSync(dir, { withFileTypes: true });\n    } catch {\n      continue;\n    }\n    for (const file of files) {\n      if (!file.isFile() || file.name.startsWith('.')) continue;\n      try {\n        mkdirSync(deadDir, { recursive: true });\n        renameSync(join(dir, file.name), join(deadDir, file.name));\n        moved++;\n      } catch {\n        // best-effort — a marker that vanished or can't move is left as-is\n      }\n    }\n  }\n  return moved;\n}\n\n/**\n * Compute the pane.log age for each agent. Missing or unreadable\n * pane.log returns null — the caller should drop those entries\n * rather than fabricate a \"fresh\" or \"ancient\" value. A missing\n * file means the agent has never spawned in this manager generation,\n * which is a separate problem covered by SessionAliveAgeSeconds.\n */\nexport function collectResponsivenessProbes(\n  codeNames: string[],\n  now: Date = new Date(),\n): ResponsivenessProbeResult[] {\n  const nowMs = now.getTime();\n  const results: ResponsivenessProbeResult[] = [];\n  for (const codeName of codeNames) {\n    try {\n      const panePath = paneLogPath(codeName);\n      const mtimeMs = statSync(panePath).mtimeMs;\n      const ageSeconds = Math.max(0, Math.floor((nowMs - mtimeMs) / 1000));\n      const result: ResponsivenessProbeResult = {\n        code_name: codeName,\n        pane_activity_age_seconds: ageSeconds,\n      };\n      // ENG-6017: piggyback the pending-inbound drain-age scan on the same\n      // cadence. Field omitted (not 0) when there are no markers.\n      const oldestMarkerMs = oldestPendingInboundMtimeMs(dirname(panePath));\n      if (oldestMarkerMs !== null) {\n        result.pending_inbound_oldest_age_seconds = Math.max(\n          0,\n          Math.floor((nowMs - oldestMarkerMs) / 1000),\n        );\n      }\n      results.push(result);\n    } catch {\n      // No pane.log yet (fresh agent, never spawned) — skip. The\n      // session-alive monitor already covers the \"should be running\n      // but isn't\" case.\n    }\n  }\n  return results;\n}\n"],"mappings":";;;;;;;AAiCA,SAAS,WAAW,aAAa,cAAc,YAAY,gBAAgB;AAC3E,SAAS,SAAS,YAAY;AAc9B,IAAM,sBAAsB,IAAI,KAAK;AAE9B,SAAS,8BAAsC;AACpD,QAAM,MAAM,QAAQ,IAAI;AACxB,MAAI,CAAC,IAAK,QAAO;AACjB,QAAM,SAAS,OAAO,SAAS,KAAK,EAAE;AACtC,SAAO,OAAO,SAAS,MAAM,KAAK,SAAS,IAAI,SAAS;AAC1D;AAoBA,SAAS,4BAA4B,cAAqC;AACxE,MAAI,SAAwB;AAC5B,MAAI;AACJ,MAAI;AACF,cAAU,YAAY,cAAc,EAAE,eAAe,KAAK,CAAC;AAAA,EAC7D,QAAQ;AACN,WAAO;AAAA,EACT;AACA,aAAW,SAAS,SAAS;AAC3B,QAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,KAAK,SAAS,kBAAkB,EAAG;AACtE,UAAM,MAAM,KAAK,cAAc,MAAM,IAAI;AACzC,QAAI;AACJ,QAAI;AACF,cAAQ,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;AAAA,IAClD,QAAQ;AACN;AAAA,IACF;AACA,eAAW,QAAQ,OAAO;AACxB,UAAI,CAAC,KAAK,OAAO,KAAK,KAAK,KAAK,WAAW,GAAG,EAAG;AACjD,UAAI;AACF,cAAM,UAAU,SAAS,KAAK,KAAK,KAAK,IAAI,CAAC,EAAE;AAC/C,YAAI,WAAW,QAAQ,UAAU,OAAQ,UAAS;AAAA,MACpD,QAAQ;AAAA,MAER;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAUA,SAAS,sBAAsB,YAAoC;AACjE,MAAI;AACF,UAAM,SAAS,KAAK,MAAM,aAAa,YAAY,MAAM,CAAC;AAC1D,WAAO,QAAQ,kBAAkB;AAAA,EACnC,SAAS,OAAO;AACd,WAAQ,MAAgC,SAAS,WAAW,OAAO;AAAA,EACrE;AACF;AAmBO,SAAS,gCACd,cACA,OAA2C,CAAC,GAC7B;AACf,QAAM,iBAAiB,KAAK,kBAAkB;AAC9C,MAAI,SAAwB;AAC5B,MAAI;AACJ,MAAI;AACF,cAAU,YAAY,cAAc,EAAE,eAAe,KAAK,CAAC;AAAA,EAC7D,QAAQ;AACN,WAAO;AAAA,EACT;AACA,aAAW,SAAS,SAAS;AAC3B,QAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,KAAK,SAAS,kBAAkB,EAAG;AACtE,UAAM,MAAM,KAAK,cAAc,MAAM,IAAI;AACzC,QAAI;AACJ,QAAI;AACF,cAAQ,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;AAAA,IAClD,QAAQ;AACN;AAAA,IACF;AACA,eAAW,QAAQ,OAAO;AACxB,UAAI,CAAC,KAAK,OAAO,KAAK,KAAK,KAAK,WAAW,GAAG,EAAG;AACjD,YAAM,OAAO,KAAK,KAAK,KAAK,IAAI;AAChC,UAAI;AACJ,UAAI;AACF,kBAAU,SAAS,IAAI,EAAE;AAAA,MAC3B,QAAQ;AACN;AAAA,MACF;AACA,UAAI,mBAAmB,QAAQ,UAAU,eAAgB;AACzD,YAAM,gBAAgB,sBAAsB,IAAI;AAChD,UAAI,kBAAkB,KAAM;AAC5B,UAAI,cAAe;AACnB,UAAI,WAAW,QAAQ,UAAU,OAAQ,UAAS;AAAA,IACpD;AAAA,EACF;AACA,SAAO;AACT;AAQO,SAAS,mCACd,UACA,gBACA,MAAY,oBAAI,KAAK,GACN;AACf,QAAM,SAAS,gCAAgC,QAAQ,YAAY,QAAQ,CAAC,GAAG,EAAE,eAAe,CAAC;AACjG,MAAI,WAAW,KAAM,QAAO;AAC5B,SAAO,KAAK,IAAI,GAAG,KAAK,OAAO,IAAI,QAAQ,IAAI,UAAU,GAAI,CAAC;AAChE;AAcO,SAAS,yBAAyB,UAAkB,OAAa,oBAAI,KAAK,GAAW;AAC1F,QAAM,OAAO,QAAQ,YAAY,QAAQ,CAAC;AAC1C,MAAI,QAAQ;AACZ,MAAI;AACJ,MAAI;AACF,cAAU,YAAY,MAAM,EAAE,eAAe,KAAK,CAAC;AAAA,EACrD,QAAQ;AACN,WAAO;AAAA,EACT;AACA,aAAW,SAAS,SAAS;AAC3B,QAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,KAAK,SAAS,kBAAkB,EAAG;AACtE,UAAM,MAAM,KAAK,MAAM,MAAM,IAAI;AACjC,UAAM,UAAU,KAAK,MAAM,GAAG,MAAM,IAAI,QAAQ;AAChD,QAAI;AACJ,QAAI;AACF,cAAQ,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;AAAA,IAClD,QAAQ;AACN;AAAA,IACF;AACA,eAAW,QAAQ,OAAO;AACxB,UAAI,CAAC,KAAK,OAAO,KAAK,KAAK,KAAK,WAAW,GAAG,EAAG;AACjD,UAAI;AACF,kBAAU,SAAS,EAAE,WAAW,KAAK,CAAC;AACtC,mBAAW,KAAK,KAAK,KAAK,IAAI,GAAG,KAAK,SAAS,KAAK,IAAI,CAAC;AACzD;AAAA,MACF,QAAQ;AAAA,MAER;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AASO,SAAS,4BACd,WACA,MAAY,oBAAI,KAAK,GACQ;AAC7B,QAAM,QAAQ,IAAI,QAAQ;AAC1B,QAAM,UAAuC,CAAC;AAC9C,aAAW,YAAY,WAAW;AAChC,QAAI;AACF,YAAM,WAAW,YAAY,QAAQ;AACrC,YAAM,UAAU,SAAS,QAAQ,EAAE;AACnC,YAAM,aAAa,KAAK,IAAI,GAAG,KAAK,OAAO,QAAQ,WAAW,GAAI,CAAC;AACnE,YAAM,SAAoC;AAAA,QACxC,WAAW;AAAA,QACX,2BAA2B;AAAA,MAC7B;AAGA,YAAM,iBAAiB,4BAA4B,QAAQ,QAAQ,CAAC;AACpE,UAAI,mBAAmB,MAAM;AAC3B,eAAO,qCAAqC,KAAK;AAAA,UAC/C;AAAA,UACA,KAAK,OAAO,QAAQ,kBAAkB,GAAI;AAAA,QAC5C;AAAA,MACF;AACA,cAAQ,KAAK,MAAM;AAAA,IACrB,QAAQ;AAAA,IAIR;AAAA,EACF;AACA,SAAO;AACT;","names":[]}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@integrity-labs/agt-cli",
-  "version": "0.27.150-test.15",
+  "version": "0.27.150-test.17",
   "description": "Augmented Team CLI — agent provisioning and management",
   "type": "module",
   "engines": {