@mmerterden/multi-agent-pipeline 14.2.1 → 14.2.2

package/pipeline/scripts/run-aggregator.mjs

@@ -82,14 +82,27 @@ function die(msg) {
   process.exit(1);
 }
 
-function resolveTaskDir() {
-  if (flags["task-dir"]) return flags["task-dir"];
-  if (flags["task-id"] && flags.project) {
+// phase-tracker.sh writes tracker-state.json (never phase-tracker.json) at
+// $HOME/.claude/logs/multi-agent/<task-id>/ (no <project> segment) - the
+// same drift render-cost-summary.sh/render-agent-log-cost.sh already guard
+// against with a candidate list. Mirror that list here instead of assuming
+// one fixed filename/path.
+function resolveTrackerCandidates() {
+  const candidates = [];
+  if (flags["task-dir"]) {
+    candidates.push(join(flags["task-dir"], "tracker-state.json"));
+  }
+  if (flags["task-id"]) {
     const HOME = process.env.HOME || process.env.USERPROFILE;
     if (!HOME) die("neither HOME nor USERPROFILE is set  -  pass --task-dir");
-    return join(HOME, ".claude", "logs", "multi-agent", flags.project, String(flags["task-id"]));
+    const taskId = String(flags["task-id"]);
+    if (flags.project) {
+      candidates.push(join(HOME, ".claude", "logs", "multi-agent", flags.project, taskId, "tracker-state.json"));
+    }
+    candidates.push(join(HOME, ".claude", "logs", "multi-agent", taskId, "tracker-state.json"));
   }
-  die("either --task-dir or --task-id with --project is required");
+  if (!candidates.length) die("either --task-dir or --task-id is required");
+  return candidates;
 }
 
 function loadJson(path) {
@@ -161,11 +174,14 @@ function tsCompare(a, b) {
 }
 
 // ─────────────────────────────────────────────────────────────────────────
-const taskDir = resolveTaskDir();
-if (!existsSync(taskDir)) die(`task dir not found: ${taskDir}`);
-
-const tracker = loadJson(join(taskDir, "phase-tracker.json"));
-if (!tracker) die(`phase-tracker.json not found in ${taskDir}`);
+const trackerCandidates = resolveTrackerCandidates();
+const trackerPath = trackerCandidates.find((p) => existsSync(p));
+if (!trackerPath) {
+  die(`tracker state not found, tried:\n${trackerCandidates.map((p) => `  ${p}`).join("\n")}`);
+}
+const taskDir = dirname(trackerPath);
+const tracker = loadJson(trackerPath);
+if (!tracker) die(`failed to parse tracker state at ${trackerPath}`);
 
 const otelSpans = loadJsonl(join(taskDir, "otel-spans.jsonl"));
 const costTable = loadCostTable();
@@ -187,7 +203,15 @@ let totalOut = 0;
 let totalCost = 0;
 let unknownModelHits = 0;
 
-for (const phase of tracker.phases || []) {
+// tracker.phases ships in two shapes (same contract render-cost-summary.sh
+// normalizes): an array (phase-tracker.sh's real output) or an object keyed
+// by phase id (older fixtures / hand-built variants). Normalize once so the
+// rest of this file can treat it as a plain array.
+const phaseList = Array.isArray(tracker.phases)
+  ? tracker.phases
+  : Object.entries(tracker.phases || {}).map(([key, value]) => ({ ...value, id: value?.id ?? key }));
+
+for (const phase of phaseList) {
   const tIn = phase.tokens_in || 0;
   const tOut = phase.tokens_out || 0;
   const model = phase.model || inferModelForPhase(phase.id);
@@ -279,9 +303,9 @@ const summary = {
   kind: "summary",
   ts: lastTs,
   task_id: taskId,
-  phases_total: tracker.phases?.length ?? 0,
-  phases_completed: (tracker.phases || []).filter((p) => p.status === "completed").length,
-  phases_failed: (tracker.phases || []).filter((p) => p.status === "failed").length,
+  phases_total: phaseList.length,
+  phases_completed: phaseList.filter((p) => p.status === "completed").length,
+  phases_failed: phaseList.filter((p) => p.status === "failed").length,
   tokens_in: totalIn,
   tokens_out: totalOut,
   total_cost_usd: Number(totalCost.toFixed(6)),

package/pipeline/scripts/smoke-schema-validation.sh

@@ -91,10 +91,14 @@ echo "→ 4. prefs.schema.json  -  keychainMapping covers every resolved service
 # A floor, not a pin: new credentials get added over time, and pinning the count
 # made every addition a false failure. What matters is that the keys shipped
 # flows resolve are all declared.
+# process.stdout.write of a String, not console.log of a Number: console.log
+# inspects non-string arguments, so with FORCE_COLOR set in the environment the
+# count comes back wrapped in ANSI codes and the -ge comparison below silently
+# fails on a value that is actually in range.
 KM_COUNT=$(node -e "
   const s = JSON.parse(require('fs').readFileSync('$PREFS_SCHEMA', 'utf8'));
   const km = s.properties.global.properties.keychainMapping.properties;
-  console.log(Object.keys(km).length);
+  process.stdout.write(String(Object.keys(km).length));
 " 2>/dev/null)
 if [ "${KM_COUNT:-0}" -ge 12 ] 2>/dev/null; then
   pass "keychainMapping declares $KM_COUNT services (≥12 expected)"

package/pipeline/scripts/test-gap-scan.mjs

@@ -141,6 +141,45 @@ function isExcluded(path) {
   return false;
 }
 
+// git quotes a `diff --git` path (core.quotePath, on by default) whenever it
+// has a non-ASCII byte or other unusual character - e.g. a Turkish filename
+// renders as `diff --git "a/\303\226deme.swift" "b/\303\226deme.swift"`, not
+// the bare a/...b/... form. The plain regex this replaced never matched that
+// line at all, so every non-ASCII-named file's added lines were silently
+// dropped from the test-gap scan.
+const DIFF_GIT_QUOTED_RE = /^diff --git "a\/((?:[^"\\]|\\.)*)" "b\/((?:[^"\\]|\\.)*)"$/;
+const DIFF_GIT_UNQUOTED_RE = /^diff --git a\/(.+?) b\/(.+)$/;
+
+function unquoteGitPath(s) {
+  const bytes = [];
+  for (let i = 0; i < s.length; i++) {
+    if (s[i] === "\\") {
+      const octal = s.slice(i + 1, i + 4);
+      if (/^[0-7]{3}$/.test(octal)) {
+        bytes.push(parseInt(octal, 8));
+        i += 3;
+        continue;
+      }
+      const simple = { "\\": 92, '"': 34, t: 9, n: 10, r: 13 };
+      const next = s[i + 1];
+      if (next in simple) {
+        bytes.push(simple[next]);
+        i += 1;
+        continue;
+      }
+    }
+    bytes.push(s.charCodeAt(i));
+  }
+  return Buffer.from(bytes).toString("utf-8");
+}
+
+function matchDiffGitHeader(line) {
+  const q = line.match(DIFF_GIT_QUOTED_RE);
+  if (q) return unquoteGitPath(q[2]);
+  const u = line.match(DIFF_GIT_UNQUOTED_RE);
+  return u ? u[2] : null;
+}
+
 function isTestPath(path) {
   if (Array.isArray(rules.testPathSuffixes) && rules.testPathSuffixes.some((s) => path.endsWith(s)))
     return true;
@@ -190,9 +229,9 @@ function parseDiffAddedLines(diffText) {
   const perFile = new Map();
   let cur = null;
   for (const line of diffText.split("\n")) {
-    const m = line.match(/^diff --git a\/(.+?) b\/(.+)$/);
-    if (m) {
-      cur = m[2];
+    const p = matchDiffGitHeader(line);
+    if (p !== null) {
+      cur = p;
       if (!perFile.has(cur)) perFile.set(cur, { lines: [], isNew: false });
       continue;
     }
@@ -300,9 +339,9 @@ function symbolMentionedInTest(diffText, symbol) {
   const lines = diffText.split("\n");
   let inTestFile = false;
   for (const line of lines) {
-    const fm = line.match(/^diff --git a\/(.+?) b\/(.+)$/);
-    if (fm) {
-      inTestFile = isTestPath(fm[2]);
+    const fm = matchDiffGitHeader(line);
+    if (fm !== null) {
+      inTestFile = isTestPath(fm);
       continue;
     }
     if (!inTestFile) continue;

package/pipeline/scripts/uninstall.mjs

@@ -248,6 +248,31 @@ function rmMatchingFiles(parent, predicate) {
   return count;
 }
 
+// Copilot CLI / Codex CLI have no plugin loader, so install/_plugin-skills.mjs
+// copies each stack plugin's authored skills straight into the flat skills/
+// dir (~100 dirs across a typical set of enabled plugins). None of those
+// names are in PIPELINE_CORE_SKILL_DIRS - the allowlist above only covers the
+// pipeline's OWN skills - so they were never removed on uninstall. The
+// installer now writes PLUGIN_SKILLS_MANIFEST listing exactly what it
+// delivered; read it back here rather than guessing plugin skill names.
+const PLUGIN_SKILLS_MANIFEST = ".plugin-skills-manifest.json";
+
+function rmPluginDeliveredSkills(skillsDir) {
+  const manifestPath = join(skillsDir, PLUGIN_SKILLS_MANIFEST);
+  if (!existsSync(manifestPath)) return 0;
+  let names;
+  try {
+    names = JSON.parse(readFileSync(manifestPath, "utf-8"));
+  } catch {
+    return 0;
+  }
+  if (!Array.isArray(names)) return 0;
+  const nameSet = new Set(names);
+  const count = rmMatchingDirs(skillsDir, (name) => nameSet.has(name));
+  rmIfExists(manifestPath);
+  return count;
+}
+
 // Explicit terminator install/copilot.mjs writes after the pipeline section
 // (v11.4.1+). Must stay in sync with INSTRUCTIONS_END_MARKER there.
 const COPILOT_END_MARKER = "<!-- multi-agent-pipeline:copilot-instructions:end -->";
@@ -387,13 +412,20 @@ function cleanClaudeSettings(settingsPath) {
   if (settings.hooks?.PreToolUse?.length) {
     const before = settings.hooks.PreToolUse.length;
     // Match both the current matcher ("Bash") and the dead pre-v11.4.1 one
-    // ("Bash(git commit:*)"); the pre-commit-check.sh command identifies the
-    // entry as ours either way.
+    // ("Bash(git commit:*)"); the command identifies the entry as ours either
+    // way. install/claude.mjs registers TWO PreToolUse gates this way -
+    // pre-commit-check.sh AND agent-guard.sh - and only the first was ever
+    // stripped here. After ~/.claude/scripts/ is deleted below, the
+    // agent-guard.sh entry pointed at a script that no longer existed, so
+    // every subsequent Bash call fired a hook error.
     settings.hooks.PreToolUse = settings.hooks.PreToolUse.filter(
       (h) =>
         !(
           (h.matcher === "Bash" || h.matcher === "Bash(git commit:*)") &&
-          h.hooks?.some((sub) => sub.command?.includes("pre-commit-check.sh"))
+          h.hooks?.some(
+            (sub) =>
+              sub.command?.includes("pre-commit-check.sh") || sub.command?.includes("agent-guard.sh"),
+          )
         ),
     );
     if (settings.hooks.PreToolUse.length < before) touched = true;
@@ -540,10 +572,12 @@ export async function main() {
         name === "figma-to-component",
     );
     n += rmMatchingDirs(skills, (name) => PIPELINE_CORE_SKILL_DIRS.includes(name));
+    n += rmPluginDeliveredSkills(skills);
     if (n > 0) console.log(`    removed ${n} skill dir(s) under ${skills}`);
     rmIfExists(join(skills, ".skills-index.json"));
     rmIfExists(join(skills, "skills-index.md"));
     rmIfExists(join(skills, "README.md"));
+    rmIfExists(join(COP, "rules"));
     // Agent definitions are FILES; remove only the pipeline-shipped names so
     // user-authored agents survive.
     rmMatchingFiles(join(COP, "agents"), (name) => PIPELINE_AGENT_FILES.includes(name));
@@ -613,10 +647,11 @@ export async function main() {
     // Codex's own `.system` set plus user skills; `prompts/` and `agents/` hold
     // user-authored files.
     const skills = join(CODEX, "skills");
-    const n = rmMatchingDirs(
+    let n = rmMatchingDirs(
       skills,
       (name) => name === "multi-agent" || name.startsWith("multi-agent-"),
     );
+    n += rmPluginDeliveredSkills(skills);
     if (n > 0) console.log(`    removed ${n} skill dir(s) under ${skills}`);
     rmIfExists(join(CODEX, "prompts", "multi-agent.md"));
     removeGeneratedCodexAgents(join(CODEX, "agents"));

package/pipeline/scripts/update-issue-progress.sh

@@ -14,8 +14,8 @@
 #   update-issue-progress.sh <task-id>
 #
 # Inputs:
-#   $HOME/.claude/projects/{slug}/state/{taskId}/agent-state.json
-#     .issue.url, .issue.number, .repos.primary.org, .repos.primary.name
+#   $HOME/.claude/logs/multi-agent/{project}/{taskId}/agent-state.json
+#     .githubIssue.owner, .githubIssue.repo, .githubIssue.number
 #     .flags.implementation, .flags.testing, .flags.codeConnect, .flags.wiki
 #
 # Side effects:
@@ -86,11 +86,12 @@ if [ "$TASK_ID" = "--rewrite" ]; then
   exit 0
 fi
 
-# Resolve the project state directory (matches phase-tracker convention).
-STATE_GLOB="$HOME/.claude/projects/*"
+# Resolve agent-state.json. multi-repo-pipeline.sh writes it under a
+# per-project directory ($HOME/.claude/logs/multi-agent/{project}/{taskId}/)
+# - not the ~/.claude/projects/*/state/ path this used to glob, which no
+# writer in the pipeline ever populates.
 AGENT_STATE=""
-for dir in $STATE_GLOB; do
-  candidate="$dir/state/$TASK_ID/agent-state.json"
+for candidate in "$HOME"/.claude/logs/multi-agent/*/"$TASK_ID"/agent-state.json; do
   if [ -f "$candidate" ]; then
     AGENT_STATE="$candidate"
     break
@@ -102,18 +103,13 @@ if [ -z "$AGENT_STATE" ]; then
   exit 2
 fi
 
-ISSUE_URL=$(jq -r '.issue.url // empty' "$AGENT_STATE")
-ISSUE_NUM=$(jq -r '.issue.number // empty' "$AGENT_STATE")
-ORG_REPO=$(jq -r '.repos.primary.name // empty' "$AGENT_STATE")
-# Derive org/repo from URL if not present in state.
-if [ -z "$ORG_REPO" ] || ! [[ "$ORG_REPO" == */* ]]; then
-  if [[ "$ISSUE_URL" =~ github\.com/([^/]+/[^/]+)/issues/ ]]; then
-    ORG_REPO="${BASH_REMATCH[1]}"
-  fi
-fi
+# agent-state.schema.json has no `.issue`/`.repos` properties (additionalProperties
+# is false) - the real GitHub-issue coordinates live at `.githubIssue.{owner,repo,number}`.
+ISSUE_NUM=$(jq -r '.githubIssue.number // empty' "$AGENT_STATE")
+ORG_REPO=$(jq -r 'if .githubIssue then (.githubIssue.owner + "/" + .githubIssue.repo) else empty end' "$AGENT_STATE")
 
 if [ -z "$ORG_REPO" ] || [ -z "$ISSUE_NUM" ]; then
-  echo "update-issue-progress: cannot resolve org/repo or issue number from state" >&2
+  echo "update-issue-progress: no githubIssue in state for task=$TASK_ID (Jira-originated task, or issue not yet linked)" >&2
   exit 3
 fi
 

package/pipeline/scripts/worktree-finalize.sh

@@ -202,6 +202,7 @@ fi
 # copied only if present; a task that never reached Phase 4 has no triage output
 # and that is not an error.
 DEST="$LOGS_ROOT/${PROJECT:+$PROJECT/}$TASK_ID/artifacts"
+LOGS_TASK_DIR="$LOGS_ROOT/${PROJECT:+$PROJECT/}$TASK_ID"
 
 copy_one() {
   src="$rp_wt/$1"
@@ -214,7 +215,27 @@ copy_one() {
   cp -R "$src" "$DEST/$1" 2>/dev/null && SALVAGED="${SALVAGED:+$SALVAGED,}\"$1\""
 }
 
-for a in agent-state.json phase-tracker.json triage-output.json \
+# agent-state.json and the tracker state file are never written INSIDE the
+# worktree checkout - multi-repo-pipeline.sh/phase-tracker.sh write both
+# straight to $LOGS_TASK_DIR. copy_one() sourcing from $rp_wt always missed
+# them (src never existed → silent no-op), so this salvage - and the
+# worktreeRemovedAt stamp below, gated on the salvaged copy existing - never
+# actually fired for either file.
+copy_from_logs() {
+  src="$LOGS_TASK_DIR/$1"
+  [ -e "$src" ] || return 0
+  if [ "$DRY" = "1" ]; then
+    SALVAGED="${SALVAGED:+$SALVAGED,}\"$1\""
+    return 0
+  fi
+  mkdir -p "$DEST/$(dirname "$1")" 2>/dev/null
+  cp -R "$src" "$DEST/$1" 2>/dev/null && SALVAGED="${SALVAGED:+$SALVAGED,}\"$1\""
+}
+
+copy_from_logs agent-state.json
+copy_from_logs tracker-state.json
+
+for a in triage-output.json \
          .review-diff.txt .build.log .test.log .pipeline; do
   copy_one "$a"
 done
@@ -230,7 +251,7 @@ fi
 # tree. This is what keeps `--force` off the table: a refusal after this point
 # means unexpected dirt, and forcing past unexpected dirt is how work gets lost.
 DROPPED=""
-for a in agent-state.json phase-tracker.json triage-output.json \
+for a in triage-output.json \
          .review-diff.txt .build.log .test.log .pipeline; do
   [ -e "$rp_wt/$a" ] || continue
   case "$a" in

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

@@ -1,6 +1,6 @@
 {
   "schemaVersion": "1.0.0",
-  "skillCount": 201,
+  "skillCount": 202,
   "entries": [
     {
       "name": "accessibility-compliance-accessibility-audit",
@@ -695,6 +695,15 @@
       "triggerPaths": [],
       "relativePath": "shared/external/live-activities/SKILL.md"
     },
+    {
+      "name": "localization-reuse-map",
+      "description": ">-",
+      "platform": null,
+      "group": "external",
+      "triggerKeywords": [],
+      "triggerPaths": [],
+      "relativePath": "shared/external/localization-reuse-map/SKILL.md"
+    },
     {
       "name": "macos-menubar-tuist-app",
       "description": "Build, refactor, or review SwiftUI macOS menubar apps that use Tuist. Use when building, refactoring or reviewing a SwiftUI macOS menubar app that uses Tuist.",
@@ -922,7 +931,7 @@
     },
     {
       "name": "multi-agent-language",
-      "description": "Toggle outputLanguage (assistant explanations). promptLanguage is fixed to English. External payloads stay English. Use when the assistant should explain itself in a different language.",
+      "description": "Toggle outputLanguage (assistant explanations, picker questions, PR/Jira/Confluence bodies). promptLanguage is fixed to English; commit messages, branch names and identifiers stay English. Use when the assistant should explain itself in a different language.",
       "platform": null,
       "group": "core",
       "triggerKeywords": [],

package/pipeline/skills/shared/README.md

@@ -47,7 +47,7 @@ Source layout is logical grouping only  -  skill discovery at runtime is unchang
 | [`multi-agent-issue`](./core/multi-agent-issue/) | `core` | List unassigned GitHub issues, pick one, auto-assign, and launch the multi-agent pipeline. Use when a GitHub issue should be picked up and s |
 | [`multi-agent-jira`](./core/multi-agent-jira/) | `core` | List open Jira issues, pick one, and launch the multi-agent pipeline. Use when a Jira issue should be picked up and started without knowing  |
 | [`multi-agent-kill`](./core/multi-agent-kill/) | `core` | Stop the given task, then remove its worktree and branch. Asks for confirmation. Use when a running or stuck task should be stopped and its  |
-| [`multi-agent-language`](./core/multi-agent-language/) | `core` | Toggle outputLanguage (assistant explanations). promptLanguage is fixed to English. External payloads stay English. Use when the assistant s |
+| [`multi-agent-language`](./core/multi-agent-language/) | `core` | Toggle outputLanguage (assistant explanations, picker questions, PR/Jira/Confluence bodies). promptLanguage is fixed to English; commit mess |
 | [`multi-agent-local`](./core/multi-agent-local/) | `core` | Full pipeline in local mode  -  no worktree, runs directly on the current branch. Use when the full pipeline should run on the current branc |
 | [`multi-agent-local-autopilot`](./core/multi-agent-local-autopilot/) | `core` | Full pipeline + local + autopilot  -  no worktree, no confirmations, all 8 phases run end-to-end on the current branch. Use when the full pi |
 | [`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. |

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

@@ -49,3 +49,7 @@ Run the task end-to-end with no confirmations.
 2. **Set the autopilot flag**  -  `"autopilot": true` in `agent-state.json`
 3. **Start the multi-agent pipeline**  -  All confirmations are skipped
 4. **On failure**  -  After 3 retries still failing → pause, ask the user
+
+## Required: outward-facing payload contracts
+
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.

package/pipeline/skills/shared/core/multi-agent-create-jira/SKILL.md

@@ -12,7 +12,7 @@ argument-hint: "[\"<free-text description>\"] [figma-url] [swagger-url]  -  all
 
 Creates exactly one Jira issue, and only after explicit approval. The issue type (Task / Bug / Story) is asked at the start of every run. No branches, no commits, no worktrees. The description comes from the type's standard template (baseline, not a mined heading set); sections auto-size  -  a conditional section renders only when its trigger is present, otherwise the heading is omitted.
 
-> **Language**: issue content (summary + description) and user-facing questions follow `prefs.global.outputLanguage`  -  intentional exception to the "external payloads stay English" default (the issue is authored for the user's team). Code identifiers and URLs stay verbatim.
+> **Language**: issue content (summary + description) and user-facing questions follow `prefs.global.outputLanguage`, like every other user-facing payload body (`rules.md` matrix). Code identifiers and URLs stay verbatim.
 
 ## Standard templates (A = always, C = conditional)
 

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

@@ -64,23 +64,10 @@ Phase 7: Report     → Channels (Jira / Confluence / PR / Wiki)
 | Phase 7 channels (Jira / Confluence / PR / Wiki) | ✅ | ✅ (same) |
 | Duration | ~10-15 min | ~7-10 min |
 
-## Analysis doc supplied to a fast mode (warn before starting)
+## Required: outward-facing payload contracts
 
-The `--dev` family skips Phase 1 (Analysis) and Phase 2 (Planning) by design. So when
-the input references an analysis document  -  a Confluence URL, a local analysis file,
-or the user says "I ran analysis for this"  -  there is **no phase that turns it into a
-plan**. The doc becomes raw context for one Dev pass, and work comes out ordered by
-whatever the model read first: the bottom of the dependency chain lands, the screen
-wiring does not.
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.
 
-Say so before starting, once, and offer the choice:
+## Intake warnings (`--dev` family)
 
-```
-This mode skips Analysis and Planning, so the analysis document will not be turned
-into a task breakdown. For analysis-driven screen work, /multi-agent or
-/multi-agent:local run both phases.
-  1. Continue with --dev (doc as context only)
-  2. Switch to the full pipeline
-```
-
-Autopilot picks 1 and logs the warning rather than asking.
+Two checks belong at the top of every `--dev` run and are specified once in `$HOME/.claude/multi-agent-refs/phases/modes.md` "Intake warnings shared by the whole `--dev` family": an analysis document supplied to a mode that skips Analysis and Planning, and a branch that already carries the work (which wants `/multi-agent:ship`, not a second Dev pass). Read that section rather than reasoning about it from scratch.

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

@@ -55,3 +55,11 @@ Phase 7: Report     → Short terminal summary
 | Confirmation (test/commit/PR) | Yes | Yes | No | **No** |
 | Review | Parallel+triage (CLI-aware) | Parallel+triage (CLI-aware) | Parallel+triage (CLI-aware) | **Parallel+triage, auto-fix** |
 | Estimated duration | ~12 min | ~7 min | ~10 min | **~5 min** |
+
+## Required: outward-facing payload contracts
+
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.
+
+## Intake warnings (`--dev` family)
+
+Two checks belong at the top of every `--dev` run and are specified once in `$HOME/.claude/multi-agent-refs/phases/modes.md` "Intake warnings shared by the whole `--dev` family": an analysis document supplied to a mode that skips Analysis and Planning, and a branch that already carries the work (which wants `/multi-agent:ship`, not a second Dev pass). Read that section rather than reasoning about it from scratch.

package/pipeline/skills/shared/core/multi-agent-dev-local/SKILL.md

@@ -21,7 +21,7 @@ Phase 7: Report     → Jira/Wiki + log + knowledge/memory
 
 ## When to use it
 
-- Quick bug fix or small feature  -  review overhead is unnecessary
+- Quick bug fix or small feature  -  analysis and planning buy nothing here
 - You're working on the current branch and don't want to switch worktrees
 - Quick bug fix or small feature  -  analysis and planning buy nothing here
 
@@ -36,23 +36,10 @@ multi-agent-dev-local "PROJ-12345"                             # Jira
 multi-agent-dev-local "Bug: LoginView dark mode"               # Free-text
 ```
 
-## Analysis doc supplied to a fast mode (warn before starting)
-
-The `--dev` family skips Phase 1 (Analysis) and Phase 2 (Planning) by design. So when
-the input references an analysis document  -  a Confluence URL, a local analysis file,
-or the user says "I ran analysis for this"  -  there is **no phase that turns it into a
-plan**. The doc becomes raw context for one Dev pass, and work comes out ordered by
-whatever the model read first: the bottom of the dependency chain lands, the screen
-wiring does not.
+## Required: outward-facing payload contracts
 
-Say so before starting, once, and offer the choice:
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.
 
-```
-This mode skips Analysis and Planning, so the analysis document will not be turned
-into a task breakdown. For analysis-driven screen work, /multi-agent or
-/multi-agent:local run both phases.
-  1. Continue with --dev (doc as context only)
-  2. Switch to the full pipeline
-```
+## Intake warnings (`--dev` family)
 
-Autopilot picks 1 and logs the warning rather than asking.
+Two checks belong at the top of every `--dev` run and are specified once in `$HOME/.claude/multi-agent-refs/phases/modes.md` "Intake warnings shared by the whole `--dev` family": an analysis document supplied to a mode that skips Analysis and Planning, and a branch that already carries the work (which wants `/multi-agent:ship`, not a second Dev pass). Read that section rather than reasoning about it from scratch.

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

@@ -41,3 +41,11 @@ Routes to the orchestrator with the `--dev --local autopilot` flags. The pipelin
 multi-agent-dev-local-autopilot "PROJ-12345"                   # Jira
 multi-agent-dev-local-autopilot "Add retry to HomeNetworking"  # Free-text
 ```
+
+## Required: outward-facing payload contracts
+
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.
+
+## Intake warnings (`--dev` family)
+
+Two checks belong at the top of every `--dev` run and are specified once in `$HOME/.claude/multi-agent-refs/phases/modes.md` "Intake warnings shared by the whole `--dev` family": an analysis document supplied to a mode that skips Analysis and Planning, and a branch that already carries the work (which wants `/multi-agent:ship`, not a second Dev pass). Read that section rather than reasoning about it from scratch.

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

@@ -1,7 +1,7 @@
 ---
 name: multi-agent-language
 language: en
-description: "Toggle outputLanguage (assistant explanations). promptLanguage is fixed to English. External payloads stay English. Use when the assistant should explain itself in a different language."
+description: "Toggle outputLanguage (assistant explanations, picker questions, PR/Jira/Confluence bodies). promptLanguage is fixed to English; commit messages, branch names and identifiers stay English. Use when the assistant should explain itself in a different language."
 user-invocable: true
 argument-hint: "[en|tr]  -  sets outputLanguage; omit for interactive picker; use 'output en|tr' for explicit form"
 ---

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

@@ -35,3 +35,7 @@ multi-agent-local "PROJ-12345"                                 # Jira
 multi-agent-local "#42"                                        # GitHub issue
 multi-agent-local "LoginView dark mode fix"                    # Free-text
 ```
+
+## Required: outward-facing payload contracts
+
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.

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

@@ -51,3 +51,7 @@ multi-agent-local-autopilot "#3"
 ## Delegation
 
 The orchestrator skill (`multi-agent/SKILL.md`) takes the `--local` + `"autopilot": true` state flags together. Contract: `refs/phases/phase-0-init.md` Step 8 (local branch) + `refs/phases/phase-2-planning.md` Step 5 (gate skip) + Step 5c (safety classifier, v7.0.0+).
+
+## Required: outward-facing payload contracts
+
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.

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

@@ -45,3 +45,7 @@ multi-agent ship autopilot          # no gate prompts: auto-fix, auto-PR, auto-c
 - Build+Test is the automated success gate (not the interactive device user-test  -  that is `multi-agent:manual-test`). If the repo has no tests, it reports "no tests present"  -  never fabricates results.
 - Commit/PR follows house rules: conventional message, `Ref: #N` (never Closes/Fixes), NO AI/bot attribution. PR opened only if one does not already exist.
 - Full phase contract lives in the Claude Code command `commands/multi-agent/ship/SKILL.md`; this skill is the Copilot-CLI counterpart.
+
+## Required: outward-facing payload contracts
+
+Before writing anything outward-facing  -  PR body, Jira comment, Confluence page, closing report  -  load `$HOME/.claude/multi-agent-refs/payload-contracts.md`. It names the canonical section set for each payload, the markup dialect per surface (PR body is Markdown, Jira is wiki markup  -  mixing them is a defect), and the token/duration numbers the closing report must carry. Improvising a payload shape from memory is the most common failure of the short modes.

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

@@ -32,7 +32,7 @@ Run all steps automatically:
 ```
 Step 1: DETECT      Compare timestamps, find stale targets
 Step 2: COPILOT     Claude Code -> Copilot CLI (instructions + 49 sub-command skills)
-Step 2b: CODEX      Claude Code -> Codex CLI (1 router skill + 44 specs as refs + 8 agent TOML)
+Step 2b: CODEX      Claude Code -> Codex CLI (1 router skill + 49 specs as refs + 8 agent TOML)
 Step 3: REPO        Claude Code -> pipeline repo (genericized, personal data scrub)
 Step 3d: DEV-TOOLKIT Companion MCP server -> detect movement, ship gates, commit + publish
 Step 4: WEBSITE     Version + phase/model counts -> {website-host} (i18n + projects.ts)
@@ -98,7 +98,7 @@ If nothing is stale -> report "All targets up to date" and stop.
 ## Codex Sync (Step 2b)
 
 This step does **not** hand-copy files. The Codex tree is a *transform* of the Claude
-tree, not a mirror: the 43 sub-command specs become reference files (Codex silently
+tree, not a mirror: the 49 sub-command specs become reference files (Codex silently
 truncates its skills block  -  see `cross-cli-contract.md` 2.6), every reference to a
 CLI-owned tree is retargeted (`agents/<persona>.md` becomes `.toml`, the dispatcher
 becomes the router skill), the 8 personas are regenerated as TOML with a model +