@mmerterden/multi-agent-pipeline 20.28.0 → 20.30.0

package/pipeline/scripts/work-items.mjs

@@ -0,0 +1,304 @@
+#!/usr/bin/env node
+/**
+ * @file work-items.mjs  -  the work list and the PR list a client shows next
+ * to the run list, built on what the pipeline already uses:
+ *
+ *   maturity <key>   lib/issue-fetcher.sh <key>, the fetch and maturity score
+ *                    /multi-agent:jira, /multi-agent:issue and Phase 0 run;
+ *                    title, type, status, score, blockers, warnings, summary
+ *                    and a description excerpt, scrubbed
+ *   prs              open pull requests on GitHub where you are asked to
+ *                    review or are the author (`gh search prs`, the queries
+ *                    /multi-agent:review lists). Bitbucket is reported as a
+ *                    source that is not implemented, never as an empty list
+ *   verdict <task>   the review verdict of a finished /multi-agent:review run
+ *                    whose every finding was triaged (accepted, rejected or
+ *                    deferred); a running review or one with a pending
+ *                    finding is refused. The preview carries findingsDigest,
+ *                    a hash of the findings the decision was made from, so a
+ *                    confirmation can be refused when they changed since:
+ *                    `preview` computes it the way lib/post-pr-review.sh does
+ *                    (approve when no accepted blocking or important finding,
+ *                    else needs-work) without posting; `post` runs
+ *                    post-pr-review.sh itself, which posts the inline comments
+ *                    and the verdict with the user's own gh or Bitbucket
+ *                    credential
+ *
+ * The scripts run as child processes without blocking the caller's event
+ * loop.
+ *
+ * The list of assigned issues is issues.mjs (`GET /v1/issues`); nothing here
+ * fetches it a second time. Credentials are only ever read by the scripts
+ * above through credential-store.sh; no value reaches this process's output.
+ *
+ * Usage:
+ *   work-items.mjs maturity <key>
+ *   work-items.mjs prs
+ *   work-items.mjs verdict <task-id> [--post]
+ *
+ * MA_GH_BIN stands in for gh, MA_ISSUE_FETCHER and MA_POST_PR_REVIEW for the
+ * two scripts (tests).
+ *
+ * Exit: 0 printed, 1 refused, 2 usage, 3 no such run or issue.
+ *
+ * @module pipeline/scripts/work-items
+ */
+
+import { createHash } from "node:crypto";
+import { existsSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { scrub } from "./_redact.mjs";
+import { selectRuns } from "./runs-index.mjs";
+import { createRedactor } from "./transcript-index.mjs";
+import { spawnAsync } from "./_spawn-async.mjs";
+import { readRegularJson } from "../lib/regular-file.mjs";
+import { runMain } from "../lib/fatal.mjs";
+import { invokedDirectly } from "../lib/invoked-directly.mjs";
+
+export const CONTRACT_VERSION = "1.0.0";
+const HERE = dirname(fileURLToPath(import.meta.url));
+/** A Jira key, `owner/repo#N` or `#N`: the structured inputs launch accepts too. */
+export const ISSUE_KEY =
+  /^(?:[A-Z][A-Z0-9_]{0,15}-[0-9]{1,9}|[A-Za-z0-9._-]{1,100}\/[A-Za-z0-9._-]{1,100}#[0-9]{1,9}|#[0-9]{1,9})$/;
+const TIMEOUT_MS = 60_000;
+
+export class WorkError extends Error {
+  constructor(code, message) {
+    super(message);
+    this.code = code;
+  }
+}
+
+const libScript = (env, name, override) =>
+  env[override] ||
+  [join(HERE, "..", "lib", name), join(HERE, "..", "..", "lib", name)].find((p) => existsSync(p));
+
+/**
+ * The maturity check for one issue, from issue-fetcher.sh.
+ *
+ * @returns {{contractVersion, key, title, type, status, kind, maturity, description}}
+ */
+export async function issueMaturity(key, env = process.env) {
+  if (typeof key !== "string" || !ISSUE_KEY.test(key)) {
+    throw new WorkError("bad-key", "the key is not a Jira key, owner/repo#N or #N");
+  }
+  const script = libScript(env, "issue-fetcher.sh", "MA_ISSUE_FETCHER");
+  if (!script) throw new WorkError("not-installed", "issue-fetcher.sh is not installed");
+  const r = await spawnAsync("bash", [script, key], { env, timeoutMs: TIMEOUT_MS });
+  let d;
+  try {
+    d = JSON.parse(r.stdout);
+  } catch {
+    throw new WorkError("fetch-failed", "the issue could not be fetched");
+  }
+  if (!d || typeof d !== "object")
+    throw new WorkError("fetch-failed", "the issue could not be fetched");
+  const R = createRedactor();
+  const m = d.maturity && typeof d.maturity === "object" ? d.maturity : {};
+  const codes = (v) => (Array.isArray(v) ? v.map((x) => String(x?.code ?? x)).slice(0, 20) : []);
+  return {
+    contractVersion: CONTRACT_VERSION,
+    key,
+    kind: ["jira", "github", "freetext"].includes(d.kind) ? d.kind : "freetext",
+    title: R.excerpt(d.title ?? "", 200),
+    type: d.type ? String(d.type) : null,
+    status: d.status ? String(d.status) : null,
+    maturity: {
+      score: Number.isFinite(m.score) ? m.score : null,
+      blockers: codes(m.blockers),
+      warnings: codes(m.warnings),
+      summary: m.summary ? R.excerpt(m.summary, 400) : null,
+    },
+    description: d.description ? R.excerpt(d.description, 2000) : null,
+  };
+}
+
+async function ghSearch(env, role) {
+  const gh = env.MA_GH_BIN || "gh";
+  const flag = role === "reviewer" ? "--review-requested=@me" : "--author=@me";
+  const r = await spawnAsync(
+    gh,
+    [
+      "search",
+      "prs",
+      "--state",
+      "open",
+      flag,
+      "--json",
+      "number,title,url,repository,author,updatedAt",
+      "--limit",
+      "50",
+    ],
+    { env, timeoutMs: TIMEOUT_MS },
+  );
+  if (r.error || r.status !== 0) return { ok: false, items: [] };
+  try {
+    const rows = JSON.parse(r.stdout);
+    return {
+      ok: true,
+      items: (Array.isArray(rows) ? rows : []).map((p) => ({
+        source: "github",
+        repo: String(p?.repository?.nameWithOwner ?? p?.repository?.name ?? ""),
+        number: Number(p?.number) || 0,
+        title: scrub(String(p?.title ?? "")),
+        url: typeof p?.url === "string" ? p.url : null,
+        author: p?.author?.login ? String(p.author.login) : null,
+        updatedAt: typeof p?.updatedAt === "string" ? p.updatedAt : null,
+        role,
+      })),
+    };
+  } catch {
+    return { ok: false, items: [] };
+  }
+}
+
+/** Open pull requests on every host the pipeline can list, one list. */
+export async function pullRequests({ env = process.env, redact = false } = {}) {
+  const [reviewer, author] = await Promise.all([
+    ghSearch(env, "reviewer"),
+    ghSearch(env, "author"),
+  ]);
+  const seen = new Set();
+  const items = [];
+  for (const p of [...reviewer.items, ...author.items]) {
+    const k = `${p.repo}#${p.number}`;
+    if (seen.has(k)) continue;
+    seen.add(k);
+    items.push(redact ? { ...p, url: null } : p);
+  }
+  items.sort((a, b) => ((a.updatedAt ?? "") < (b.updatedAt ?? "") ? 1 : -1));
+  return {
+    contractVersion: CONTRACT_VERSION,
+    items,
+    sources: [
+      reviewer.ok || author.ok
+        ? { source: "github", ok: true, count: items.length }
+        : { source: "github", ok: false, count: 0, reason: "gh could not list pull requests" },
+      { source: "bitbucket", ok: false, count: 0, reason: "not-implemented" },
+    ],
+  };
+}
+
+function stateOf(taskId) {
+  const { found, records } = selectRuns({ taskId });
+  if (!found || !records.length) throw new WorkError("no-run", "no such run");
+  const dir = records[0].dir;
+  for (const p of [join(dir, "agent-state.json"), join(dir, "artifacts", "agent-state.json")]) {
+    if (existsSync(p)) return { taskId: records[0].taskId, state: readRegularJson(p, null) ?? {} };
+  }
+  throw new WorkError("no-review", "the run has no review to publish");
+}
+
+const FINISHED = new Set(["complete", "completed", "done"]);
+const TRIAGED = new Set(["accepted", "rejected", "deferred"]);
+
+/** A hash of the findings a decision is made from, in their recorded order. */
+export function findingsDigest(findings) {
+  const rows = findings.map((f) => [
+    f?.status ?? null,
+    f?.severity ?? null,
+    f?.file ?? null,
+    Number.isInteger(f?.line) ? f.line : null,
+    f?.issue ?? f?.title ?? null,
+  ]);
+  return createHash("sha256").update(JSON.stringify(rows)).digest("hex");
+}
+
+/**
+ * The verdict post-pr-review.sh would post for a review run, without posting.
+ * Refused when the run has no review input, its findings were never written,
+ * the run has not finished, or any finding is still untriaged.
+ */
+export function reviewVerdictPreview(taskId) {
+  const { taskId: id, state } = stateOf(taskId);
+  const review = state.review;
+  if (!review || typeof review !== "object" || !review.input) {
+    throw new WorkError("no-review", "the run has no review to publish");
+  }
+  if (!Array.isArray(review.findings)) {
+    throw new WorkError("no-findings", "the review has not recorded its findings yet");
+  }
+  if (!FINISHED.has(state.status)) {
+    throw new WorkError("review-incomplete", "the review run has not finished");
+  }
+  const untriaged = review.findings.filter((f) => !TRIAGED.has(f?.status)).length;
+  if (untriaged) {
+    throw new WorkError(
+      "review-incomplete",
+      `${untriaged} finding(s) are not triaged yet (accepted, rejected or deferred)`,
+    );
+  }
+  const accepted = review.findings.filter((f) => f?.status === "accepted");
+  const blocking = accepted.filter((f) => f?.severity === "blocking").length;
+  const important = accepted.filter((f) => f?.severity === "important").length;
+  return {
+    contractVersion: CONTRACT_VERSION,
+    taskId: id,
+    provider: ["github", "bitbucket-server"].includes(review.input.provider)
+      ? review.input.provider
+      : null,
+    prNumber: Number.isInteger(review.input.prNumber) ? review.input.prNumber : null,
+    decision: blocking === 0 && important === 0 ? "approve" : "needs-work",
+    acceptedBlocking: blocking,
+    acceptedImportant: important,
+    findings: review.findings.length,
+    findingsDigest: findingsDigest(review.findings),
+  };
+}
+
+/**
+ * Post the review and its verdict through post-pr-review.sh, as
+ * /multi-agent:review does. With `expected` ({decision, findingsDigest} from
+ * the preview a person confirmed), refused with `changed` when either differs
+ * now.
+ */
+export async function postReviewVerdict(taskId, env = process.env, expected = null) {
+  const preview = reviewVerdictPreview(taskId);
+  if (
+    expected &&
+    (expected.decision !== preview.decision || expected.findingsDigest !== preview.findingsDigest)
+  ) {
+    throw new WorkError("changed", "the findings or the decision changed since the preview");
+  }
+  const script = libScript(env, "post-pr-review.sh", "MA_POST_PR_REVIEW");
+  if (!script) throw new WorkError("not-installed", "post-pr-review.sh is not installed");
+  const r = await spawnAsync("bash", [script, preview.taskId], {
+    env,
+    timeoutMs: 5 * TIMEOUT_MS,
+    maxBytes: 1024 * 1024,
+  });
+  if (r.status !== 0) {
+    throw new WorkError("post-failed", `post-pr-review.sh exited ${r.status ?? "on a signal"}`);
+  }
+  return { ...preview, posted: true };
+}
+
+async function main(argv) {
+  const [cmd, arg, flag] = argv;
+  try {
+    let out;
+    if (cmd === "maturity" && arg) out = await issueMaturity(arg);
+    else if (cmd === "prs" && !arg) out = await pullRequests();
+    else if (cmd === "verdict" && arg && (flag === undefined || flag === "--post")) {
+      out = flag ? await postReviewVerdict(arg) : reviewVerdictPreview(arg);
+    } else {
+      process.stderr.write(
+        "usage: work-items.mjs maturity <key> | prs | verdict <task-id> [--post]\n",
+      );
+      return 2;
+    }
+    process.stdout.write(JSON.stringify(out, null, 2) + "\n");
+    return 0;
+  } catch (err) {
+    if (!(err instanceof WorkError)) throw err;
+    process.stderr.write(`work-items: ${err.message}\n`);
+    return err.code === "no-run" ? 3 : 1;
+  }
+}
+
+if (invokedDirectly(import.meta.url)) {
+  runMain("work-items", async () => {
+    process.exitCode = await main(process.argv.slice(2));
+  });
+}

package/pipeline/scripts/write-state.mjs

@@ -32,6 +32,9 @@
  *   4  -  the lock was taken from us mid-write; nothing was written, retry
  *   3  -  I/O error (disk full, permission denied, parent dir missing)
  *   5  -  --if-rev did not match; the record moved, nothing was written
+ *   6  -  the write moves the run into `status: complete` without the evidence
+ *         lib/run-completion.mjs requires (last phase of the mode, Phase 4
+ *         evidence, a plan-coverage verdict); nothing was written
  *
  * Tunables (env):
  *   WRITE_STATE_LOCK_TIMEOUT_MS  acquire window before exit 2 (default 15000)
@@ -67,6 +70,7 @@ import { dirname, basename } from "path";
 import { Buffer } from "node:buffer";
 import { runMain } from "../lib/fatal.mjs";
 import { invokedDirectly } from "../lib/invoked-directly.mjs";
+import { completionGaps } from "../lib/run-completion.mjs";
 
 /**
  * @typedef {{ replace: boolean, path: string, ifRev: number|null }} ParsedArgs
@@ -320,6 +324,104 @@ function undeclaredKeys(payload, current) {
   );
 }
 
+const COMPLETE = new Set(["complete", "completed"]);
+const STOP_GUARD_HALT = /^\d+:stop-guard:/;
+
+/**
+ * Lift the halt run-stop-guard.mjs records after its repeat ceiling once the
+ * run advances a phase. That halt says only "the session kept stopping"; a
+ * later phase transition is the run going on, and leaving the halt in place
+ * would show a moving run as paused and make every later Stop let it through.
+ * A halt any other step recorded, or a write that sets status or haltReason
+ * itself, is left alone.
+ *
+ * @param {object} next
+ * @param {object} current
+ * @param {unknown} payload
+ */
+export function clearStopGuardHalt(next, current, payload) {
+  if (!payload || typeof payload !== "object" || Array.isArray(payload)) return;
+  if (Object.hasOwn(payload, "status") || Object.hasOwn(payload, "haltReason")) return;
+  if (typeof current?.haltReason !== "string" || !STOP_GUARD_HALT.test(current.haltReason)) return;
+  const was = Number(current.currentPhase);
+  const now = Number(payload.currentPhase);
+  if (!Number.isInteger(now) || !(now > (Number.isInteger(was) ? was : -1))) return;
+  next.haltReason = null;
+  next.status = "in_progress";
+}
+
+/**
+ * Clear a wait the run has moved past. A write that advances `currentPhase`
+ * without setting `waitingFor` itself ends any wait recorded in the earlier
+ * phase, so a clear the step forgot cannot keep the run marked as waiting
+ * (and the stop guard off) for the rest of the run. `question` is left to
+ * answer-question.mjs, which clears it together with its pendingQuestion.
+ *
+ * @param {object} next
+ * @param {object} current
+ * @param {unknown} payload
+ */
+export function clearPassedWait(next, current, payload) {
+  if (!payload || typeof payload !== "object" || Array.isArray(payload)) return;
+  if (Object.hasOwn(payload, "waitingFor")) return;
+  if (typeof current?.waitingFor !== "string" || current.waitingFor === "question") return;
+  const was = Number(current.currentPhase);
+  const now = Number(payload.currentPhase);
+  if (!Number.isInteger(now) || !(now > (Number.isInteger(was) ? was : -1))) return;
+  next.waitingFor = null;
+}
+
+/**
+ * The refusal message for a write that moves the run into `complete` without
+ * the evidence lib/run-completion.mjs asks for, or null.
+ *
+ * Only the transition is judged: a merge that sets `status: complete`, or a
+ * replace whose document is complete over one that was not. A later write to
+ * an already complete run that does not touch `status` goes through, so the
+ * report steps after the transition keep working.
+ *
+ * @param {object} next
+ * @param {unknown} payload
+ * @param {boolean} replace
+ * @param {string|null} diskStatus
+ * @returns {string|null}
+ */
+export function completionRefusal(next, payload, replace, diskStatus) {
+  if (!COMPLETE.has(next?.status)) return null;
+  const entering = replace
+    ? !COMPLETE.has(diskStatus)
+    : Boolean(payload && typeof payload === "object" && COMPLETE.has(payload.status));
+  if (!entering) return null;
+  const gaps = completionGaps(next);
+  if (!gaps.length) return null;
+  return `write-state: refusing status "${next.status}": ${gaps.join("; ")}. Nothing was written. Finish the missing step, or record a halt (status "paused" with a haltReason).`;
+}
+
+/**
+ * Bind the run to the host session that drives it.
+ *
+ * `hostSessionId` is the Claude Code session (CLAUDE_CODE_SESSION_ID in the
+ * Bash environment) that made the run's last phase transition. The Stop hook
+ * (run-stop-guard.mjs) and the run-scoped PreToolUse guards read it to find
+ * the run a hook call belongs to. Stamped only when the write creates the
+ * document or sets `currentPhase`: those are the driver's writes, while a
+ * second session editing a running task (`/multi-agent:steer`) sets neither
+ * and so does not take the run over.
+ *
+ * @param {object} next
+ * @param {unknown} payload
+ * @param {boolean} existed
+ * @param {Record<string,string|undefined>} [env]
+ */
+export function stampHostSession(next, payload, existed, env = process.env) {
+  const sid = env.CLAUDE_CODE_SESSION_ID;
+  if (typeof sid !== "string" || !sid.trim()) return;
+  const transition =
+    payload && typeof payload === "object" && Object.hasOwn(payload, "currentPhase");
+  if (existed && !transition) return;
+  next.hostSessionId = sid.trim();
+}
+
 /**
  * @param {object} base
  * @param {object} patch
@@ -397,6 +499,8 @@ async function main() {
         }
       }
       next = deepMerge(current, payload);
+      clearStopGuardHalt(next, current, payload);
+      clearPassedWait(next, current, payload);
     }
 
     // A state document is an object. Anything else cannot carry `rev`, and
@@ -415,9 +519,11 @@ async function main() {
     // would tell the next reader that nothing changed.
     let priorRev = Number.isInteger(next.rev) ? next.rev : 0;
     let diskRev = null;
+    let diskStatus = null;
     if (existsSync(path)) {
       try {
         const parsed = JSON.parse(readFileSync(path, "utf-8"));
+        diskStatus = parsed?.status ?? null;
         if (Number.isInteger(parsed?.rev)) {
           diskRev = parsed.rev;
           priorRev = Math.max(priorRev, parsed.rev);
@@ -448,7 +554,15 @@ async function main() {
       }
     }
 
+    const refusal = completionRefusal(next, payload, replace, diskStatus);
+    if (refusal) {
+      console.error(refusal);
+      releaseLock(path);
+      process.exit(6);
+    }
+
     next.rev = priorRev + 1;
+    stampHostSession(next, payload, fileExistedBefore);
 
     const unknown = undeclaredKeys(payload, replace ? {} : current);
     if (unknown.length) {

package/pipeline/skills/.skill-manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": "1.0.0",
-  "generatedAt": "2026-10-24T11:39:37Z",
-  "skillCount": 234,
+  "generatedAt": "2026-10-24T21:22:00Z",
+  "skillCount": 237,
   "entries": [
     {
       "path": "shared/core/apple-archive-compliance/SKILL.md",
@@ -29,7 +29,7 @@
     },
     {
       "path": "shared/core/multi-agent-autopilot-on/SKILL.md",
-      "sha256": "d4f23f6f7bb493a4efa2cfa252fab50b22b6506aeaf31985db4498b5f010c8c8"
+      "sha256": "7e4fbc84ed2bac8ba5fae1b9384ffbd6a40421a3df7bc1a63ef8f0f7fe6a3e55"
     },
     {
       "path": "shared/core/multi-agent-autopilot-status/SKILL.md",
@@ -127,6 +127,18 @@
       "path": "shared/core/multi-agent-model/SKILL.md",
       "sha256": "57ba710657d74176b0f19d4a97d3183df1f7b0c056eb1eef2792308914be836c"
     },
+    {
+      "path": "shared/core/multi-agent-office-off/SKILL.md",
+      "sha256": "316c7f1cd910e76096301330cac7e02f1a2b28e66cda88e9f8bb966a14553e9e"
+    },
+    {
+      "path": "shared/core/multi-agent-office-on/SKILL.md",
+      "sha256": "f1841f3092275e54b669a2701796c187ecaa16d71b0cd4311915218319d6950c"
+    },
+    {
+      "path": "shared/core/multi-agent-office-status/SKILL.md",
+      "sha256": "fba0513f8941c026172ea0ce76cbc76d88a5e813fd713d68a1377eb5a4204c86"
+    },
     {
       "path": "shared/core/multi-agent-postmortem/SKILL.md",
       "sha256": "41edd907cbaae1523eab1cd716390c9e9cd889574141cb88c33966e221b0987d"
@@ -213,7 +225,7 @@
     },
     {
       "path": "shared/core/multi-agent-serve/SKILL.md",
-      "sha256": "90e4dd7225e10e670118ed05d0c8a9e346112c4bbbdb24a7fe7c5c334a83e848"
+      "sha256": "82ba766a7b23285f1b2b3eba2bcbd671ade82b28cc094093832075970504f732"
     },
     {
       "path": "shared/core/multi-agent-setup/SKILL.md",
@@ -221,7 +233,7 @@
     },
     {
       "path": "shared/core/multi-agent-stack/SKILL.md",
-      "sha256": "ac964cd00dd6ed06512530489d867efbc51d7d7614265f9aa022679325d5defe"
+      "sha256": "0e32b95c62324bb592d10b4c2c515509998e7260f6c4aae94271683ede9ba763"
     },
     {
       "path": "shared/core/multi-agent-status/SKILL.md",
@@ -229,7 +241,7 @@
     },
     {
       "path": "shared/core/multi-agent-steer/SKILL.md",
-      "sha256": "20f332bd2fe2745da5341f8d8187211fc47c06b4b5f5031e217affee0282ddb5"
+      "sha256": "9ac0c48c3cbafd98022a04b24ef4f16723dda6a852202002231f43bbb7dd453b"
     },
     {
       "path": "shared/core/multi-agent-store-ready/SKILL.md",
@@ -237,7 +249,7 @@
     },
     {
       "path": "shared/core/multi-agent-sync/SKILL.md",
-      "sha256": "e96afa9b89a2ab6e2fe13a01158f98272c6828e5cef03c8909240766aaefd44b"
+      "sha256": "17f734fa2b84d9a4c757f9745be8ff40d3097f5de89caf55124b48439a737093"
     },
     {
       "path": "shared/core/multi-agent-test-accessibility/SKILL.md",

package/pipeline/skills/.skills-index.json

@@ -1,6 +1,6 @@
 {
   "schemaVersion": "1.0.0",
-  "skillCount": 234,
+  "skillCount": 237,
   "entries": [
     {
       "name": "accessibility-compliance-accessibility-audit",
@@ -1355,6 +1355,39 @@
       "triggerPaths": [],
       "relativePath": "shared/core/multi-agent-model/SKILL.md"
     },
+    {
+      "name": "multi-agent-office-off",
+      "description": "Close the office panel and stop its local server. Use when done with the panel.",
+      "platform": null,
+      "group": "core",
+      "plugin": null,
+      "invokeAs": "multi-agent-office-off",
+      "triggerKeywords": [],
+      "triggerPaths": [],
+      "relativePath": "shared/core/multi-agent-office-off/SKILL.md"
+    },
+    {
+      "name": "multi-agent-office-on",
+      "description": "Open the office panel: a local browser page with agents at work, their chats, cost ledger and new runs. Use when you want to watch or drive runs from a browser.",
+      "platform": null,
+      "group": "core",
+      "plugin": null,
+      "invokeAs": "multi-agent-office-on",
+      "triggerKeywords": [],
+      "triggerPaths": [],
+      "relativePath": "shared/core/multi-agent-office-on/SKILL.md"
+    },
+    {
+      "name": "multi-agent-office-status",
+      "description": "Whether the office panel is running, its URL (never the token) and its last actions. Use when unsure if the panel is open.",
+      "platform": null,
+      "group": "core",
+      "plugin": null,
+      "invokeAs": "multi-agent-office-status",
+      "triggerKeywords": [],
+      "triggerPaths": [],
+      "relativePath": "shared/core/multi-agent-office-status/SKILL.md"
+    },
     {
       "name": "multi-agent-postmortem",
       "description": "Post-mortem of a run from its session transcripts: loops, stalls, root-cause hypotheses on file:line, fixes. Read-only, redacted. Use when a run went wrong or ran far too long.",

package/pipeline/skills/shared/README.md

@@ -2,11 +2,11 @@
 
 Single source of truth for skills delivered to both Claude Code (`~/.claude/skills/`) and Copilot CLI (`~/.copilot/skills/`) by the installer.
 
-**Total:** 234 skills (69 core + 165 external). Auto-generated by `scripts/gen-skills-index.mjs`  -  do not edit by hand.
+**Total:** 237 skills (72 core + 165 external). Auto-generated by `scripts/gen-skills-index.mjs`  -  do not edit by hand.
 
 ## Directory layout
 
-- **`core/`**  -  69 `multi-agent*` orchestration skills that are pipeline-critical. Edits here are core-code changes.
+- **`core/`**  -  72 `multi-agent*` orchestration skills that are pipeline-critical. Edits here are core-code changes.
 - **`external/`**  -  165 iOS / Android / generic skills that ship into the stack plugins. Each names its origin in `metadata.source`: `multi-agent-pipeline` when written here, or the upstream and license a `NOTICE-*.md` in this directory covers.
 - Install destinations (ADR-0009): Claude Code gets NO local copy of `external/` - it loads those skills from the `multi-agent-plugins` marketplace, namespaced (`ai-<stack>-toolkit:<name>`); only the two compliance catalogs from `core/` land in `~/.claude/skills/`. Copilot CLI and Codex CLI receive a flat copy filtered to the enabled stacks. `external/` remains the single authoring source that `build-stack-plugins.mjs` publishes from.
 
@@ -14,7 +14,7 @@ Source layout is logical grouping only  -  skill discovery at runtime is unchang
 
 ## Categories
 
-- [Pipeline Orchestration](#pipeline-orchestration)  -  69
+- [Pipeline Orchestration](#pipeline-orchestration)  -  72
 - [iOS / Apple Ecosystem](#ios-apple-ecosystem)  -  94
 - [Android / Kotlin](#android-kotlin)  -  16
 - [Web](#web)  -  10
@@ -57,6 +57,9 @@ Source layout is logical grouping only  -  skill discovery at runtime is unchang
 | [`multi-agent-log`](./core/multi-agent-log/) | `core` | Show the agent-log.md for the given task. With no ID, shows the most recent task. Use when asked what a task did, or to read its log. |
 | [`multi-agent-manual-test`](./core/multi-agent-manual-test/) | `core` | Switch to the active task's branch and prepare it for manual testing in Xcode. Phase 3 user-test step, standalone (the UI Bug Hunter lives a |
 | [`multi-agent-model`](./core/multi-agent-model/) | `core` | Turn the top model rung on or off and keep the cost ledger's pricing in step with it. Use when asked to enable or disable Fable, or which mo |
+| [`multi-agent-office-off`](./core/multi-agent-office-off/) | `core` | Close the office panel and stop its local server. Use when done with the panel. |
+| [`multi-agent-office-on`](./core/multi-agent-office-on/) | `core` | Open the office panel: a local browser page with agents at work, their chats, cost ledger and new runs. Use when you want to watch or drive  |
+| [`multi-agent-office-status`](./core/multi-agent-office-status/) | `core` | Whether the office panel is running, its URL (never the token) and its last actions. Use when unsure if the panel is open. |
 | [`multi-agent-postmortem`](./core/multi-agent-postmortem/) | `core` | Post-mortem of a run from its session transcripts: loops, stalls, root-cause hypotheses on file:line, fixes. Read-only, redacted. Use when a |
 | [`multi-agent-prune-logs`](./core/multi-agent-prune-logs/) | `core` | Delete per-task project logs under ~/.claude/logs/multi-agent (filter by age/project/task). Audit trail + metrics are preserved. Dry-run fir |
 | [`multi-agent-prune-prompts`](./core/multi-agent-prune-prompts/) | `core` | Zero-base prompt review: measure the always-on instruction footprint, classify every rule block, propose keep/trial-removal/delete; applies  |

package/pipeline/skills/shared/core/multi-agent-autopilot-on/SKILL.md

@@ -162,25 +162,23 @@ Render `templates/multi-agent-autopilot.plist.template` and show the resolved
 "Configure only" is a real answer and the safe default to offer first: the queue
 can be watched with `autopilot-status` for a while before anything runs.
 
-On **Start**:
+On **Start**, load the schedule and the awake agent with the same script the
+office panel's `POST /v1/autopilot/on` calls, naming every repo selected in step 3:
 
 ```bash
-launchctl bootstrap "gui/$(id -u)" "$HOME/Library/LaunchAgents/com.multi-agent.autopilot.plist" 2>/dev/null \
-  || launchctl load "$HOME/Library/LaunchAgents/com.multi-agent.autopilot.plist"
+node "$HOME/.claude/scripts/autopilot-control.mjs" on --repo <owner/name> [--repo <owner/name>...]
 ```
 
-Then the awake agent, rendered from
-`templates/multi-agent-autopilot-awake.plist.template` with the `awake.display`
-answer saved above:
-
-```bash
-. "$HOME/.claude/lib/autopilot-state.sh"
-node "$(ma_ap_asset scripts/autopilot-awake.mjs)" install
-```
-
-A non-zero exit means launchd refused the job: say so, with its path, and say
-the Mac may sleep between ticks until it is fixed. The schedule itself stays on.
-"Configure only" installs neither.
+It keeps exactly those repos in `config.json`, probes their credentials again
+(`autopilot-arming.mjs probeCredentials`), renders
+`templates/multi-agent-autopilot.plist.template` with its seven placeholders
+(NODE, TICK, LABEL, ROOT, INTERVAL, HOME, CLAUDE), loads it with `launchctl
+bootstrap` (falling back to `load`) and installs the awake agent with the
+`awake.display` answer saved above. `scheduled: false` with probe problems means
+nothing was loaded; say so with the problems. Exit 1 with `launchd refused`
+means launchd refused the job: say so, with its path, and say the Mac may sleep
+between ticks until it is fixed. "Configure only" is `--configure-only`, which
+saves the selection and installs neither.
 
 ### 9. Report what is now true
 

package/pipeline/skills/shared/core/multi-agent-office-off/SKILL.md

@@ -0,0 +1,16 @@
+---
+name: multi-agent-office-off
+description: "Close the office panel and stop its local server. Use when done with the panel."
+user-invocable: true
+metadata:
+  language: en
+---
+
+# multi-agent-office-off
+
+1. `node "$HOME/.copilot/scripts/office-control.mjs" off --json`
+2. Report in `outputLanguage`: whether a server was `stopped`, and the
+   `remote` line as given. A panel that was already closed is not an error.
+
+Only the recorded server is signalled, and only when its token file, command
+and start time still match.