@mmerterden/multi-agent-pipeline 20.16.0 → 20.17.1

package/CHANGELOG.md

@@ -16,6 +16,48 @@ Internal file-layout changes that don't affect the slash-command surface are sti
 
 ## [Unreleased]
 
+## [20.17.1] - 2026-10-07
+
+### Changed
+
+- Operational reporting finds the person's own GitHub account with no question asked. `usage-identity.mjs` stops at the first account that can read `usageLog.accessRepo`: a gh account as before, then the token git's credential helper holds for https://github.com (`git credential fill`, prompts disabled; login from `GET /user`, access from `GET /repos/<accessRepo>`, through gh or straight to the API when gh is missing), then an SSH key (`ssh -G github.com` identity files plus `~/.ssh/id_ed25519`, `id_ecdsa`, `id_rsa` with a `.pub`; login from `ssh -T`, access from `git ls-remote`, batch mode with only that key). Prefs cache `usageLog.loginMethod` (`gh`, `credential`, `ssh`) and `loginKey` next to the login. No token reaches an argv, a log or prefs, and a missing binary or a failed check only means no candidate.
+- `usage-register.mjs` proves the login the way it was found: a gh token or the credential helper's token (read again at registration) as the bearer, or for an SSH key a signature over a server challenge (`POST <register>/challenge`, `ssh-keygen -Y sign` over the exact challenge bytes in a 0600 temp file, stdin closed, then the `.pub` through ssh-agent), sent with no Authorization header. `401` `ssh_signature_invalid`, `ssh_key_not_on_account` and `ssh_key_unsupported`, a key that cannot sign and an endpoint without a challenge each get their own status line.
+- `needsPersonalAccount` is true only when the machine is not pinned and no account was found at all; setup and update offer `gh auth login` or adding the SSH key to the personal account as the last resort.
+
+### Removed
+
+- The fallback to the active gh account when no account can read the pipeline repository: the server accepts only the repository owner and its collaborators, so nothing is reported under another account. A cached `usageLog.loginFallback` is re-checked at once and dropped; a registration with `registeredFallback: true` is not treated as pinned, is not used to name runs, and is replaced as soon as an account with access is found.
+
+## [20.17.0] - 2026-10-07
+
+### Added
+
+- Install and update records for operational reporting (`pipeline/lib/usage-lifecycle.mjs`): the installed version is read from the installer's `.pipeline-version` stamp, its time from the stamp's modification time, and compared with `usageLog.reportedVersion`. When they differ, `usage-report.mjs --lifecycle` sends an `install` or `update` record (new version, the version it replaced, the host trees, the time) under the reporting login and moves `reportedVersion` forward once the server accepted it. Registration, setup and update with a token already onboarded, and every accepted run event run it; the installer still writes nothing but files. `usage-register.mjs` records `usageLog.registeredAt`, which tells a first install from an update on a machine that never reported a version.
+- A registration is pinned: the machine keeps the verified login in `usageLog.registeredAs` (`registeredFallback` says whether it was a fallback account), every record is named after it without a gh call, and a fallback account never replaces it. `usage-register.mjs --json` reports `needsPersonalAccount`, and setup and update offer once to log in the account that has access, switch the work account back and re-register, so the personal account needs to be logged in to gh only at that moment.
+- Valven Development Behaviors in `valven-rules` (registry 1.1.0), scored on Valven's own bands: `VLV-CYCLE-01` coding time (first commit to the pull request opening; 30 min / 2.5 h / 24 h), `VLV-CYCLE-02` pickup time (ready for review to the first review; 1 h / 3 h / 14 h) and `VLV-CYCLE-03` review time (first review to approval or merge; 30 min / 3 h / 18 h). A duration in the 50 or 0 band is a `suggestion`, because the time has passed and the pull request cannot change it. `VLV-PICKUP-01` is `important` for a pull request without a reviewer and a `suggestion` for a draft older than 3 hours (`draft_severity`), because an unattended run opens a draft on purpose; a draft is not asked for a reviewer. `VLV-REWORK-01` also reports the churn score, `100 - churn %`.
+- `valven-gate.mjs --pr-meta-file`: the pull request's timeline, read from `gh pr view --json author,closedAt,createdAt,isDraft,mergedAt,reviewRequests,reviews` as it is or from a normalized Bitbucket file; reviews by the author never count. `pr-timeline.mjs` writes the GitHub file: `gh pr view` plus the ready-for-review time and each review author's type from GraphQL, so a pull request opened as a draft is timed from when it was marked ready and a bot reviewer is not counted as a person. Pending reviews, bot reviews and reviews left before ready are ignored, an unknown reviewer count stays unknown, and a merged or closed pull request stops the clocks. The report carries `scores` per rule and `results[].score`; an unreadable file is reported and only the timeline rules go `not-measured`. `/multi-agent:review` passes it for GitHub and Bitbucket pull requests and prints a `Valven scores` line.
+- `pipeline/lib/plugin-skill-roots.mjs`: the marketplace plugin `skills/` roots, shared by `skill-conformance.mjs` and `valven-gate.mjs`.
+- Exploratory testing contract for `/multi-agent:test` (`features/exploratory-testing.md`, `explore-findings.mjs`, `explore-run.schema.json`). A finding carries expected, observed, repro steps, severity 1-5, kind `issue` or `warning`, and evidence on disk; a finding that misses a field, or writes a secret in clear text instead of `secret_ref:<name>`, cannot fail the run. Each run has a step, minute and token budget (defaults 8 / 10 / 150000, clamped to 1-12 / 3-15 / 20000-400000), and `check` stops it at a spent budget or after three consecutive blocked steps that reported nothing. `CREDENTIAL_MISSING`, `ENVIRONMENT_UNAVAILABLE`, `SEED_DATA_MISSING` and locator or replay failures are blocked steps, reported by cause and apart from product failures; a run whose every step was blocked is `blocked`, not `passed`. `sim-test.md` writes the run file, prefers `ui_locate` / `*_tap_element` and `expect_*` steps over coordinates, and renders its report from `report`.
+- Assertion judge (`features/assertion-judge.md`): a visual or semantic assertion is decided by a fresh subagent that receives only the assertion, the current screenshot or UI tree, and optionally the app's vocabulary. `explore-findings.mjs judge-packet` refuses any field describing the run, and `judge-verdict` fails an inconclusive answer, or one citing evidence it was not given, as `ASSERTION_INCONCLUSIVE`.
+- `/multi-agent:bug-bash [scope] [--platform ios|android|web] [--charters <file>]`: 5-10 charters, each with a persona (`first-time-user`, `skeptic`, `fuzzer`, `state-keeper`, `error-paths`, or `custom`) and its own budget, run as parallel explorer subagents (at most 4, one device or web `session` each). `bug-bash.mjs` validates the charters, merges findings that describe one defect, and confirms a candidate only when its repro test failed with `ASSERTION_FAILED` on the assertion that encodes it; every other candidate is listed as rejected with its reason (artifact, environment, design, fixture, did-not-reproduce, unproven, unverified), and environment rejections can be kept as unverified production risks. No commit, push or issue.
+- `e2e-testing` knowledge skill, shipped in ai-common-toolkit: element locators, expect steps and their verdict codes, `secret_ref`, failure traces, `save_as` / `replay`, parallel web sessions, and `document_generate` / `document_mrz_build` fixtures for MRZ and document-scan tests. The toolkit side of all three entries needs `@mmerterden/multi-agent-toolkit-mcp` ≥ v3.19.0 (`cross-cli-contract.md` section 1.4); on an older server the flows fall back to UI-tree reads and frame-centre taps.
+- The description-surface ceiling in `smoke-context-budget.sh` moves from 91000 to 91500 for `/multi-agent:bug-bash` and its shared/core counterpart (211 bytes each, below the 317-byte average, after a trim from 391) and the `e2e-testing` skill: the tree grew, no description is padded.
+
+### Changed
+
+- `valven-rules` moved from `skills/shared/core` to `skills/shared/external` and ships in `ai-common-toolkit`, so the rules reach everyone who installs the marketplace, not only pipeline users. Claude Code reads it from the plugin and no longer installs a local copy; Copilot CLI and Codex receive it from the installer as before. `valven-gate.mjs` and `valven-split.mjs` read the registry installed with them, so the gate's evaluators and rules come from one release: on Claude Code the installer lays the skill down in `multi-agent-refs/registries/valven-rules/`, outside the skills dir, and a leftover `~/.claude/skills/valven-rules` is never read. The plugin copies (checkout, then every cached version, newest first) are the last resort. On Claude Code `skill-conformance.mjs` searches the plugin roots first and then that registries dir; on Copilot and Codex the installed skills dir stays first.
+- The Claude Code installer removes the old local copy when it matches a shipped version: `gen-catalog-history.mjs` keys a skill by its directory name whether it sits in `core/` or `external/`. A dangling symlink a `--link` install left there is removed too. Uninstall removes the old copy and such links (`PIPELINE_RETIRED_SKILL_DIRS`).
+- The repo sync leaves `~/.claude/multi-agent-refs/registries/` out: the installer generates it from `pipeline/skills/shared/external`, which the repo already holds.
+- Phase 4 recommends opening the pull request READY rather than DRAFT: every gate has passed before the push, and a draft holds the reviewers back. A saved `defaultPrMode` still wins. The Step 2.95 log line carries the coding score.
+
+### Fixed
+
+- `pre-commit-check.sh` lets `.env.example`, `.env.sample` and `.env.template` through the env-file block, since a template documents names rather than values; their content is still scanned for secrets, and every other `.env*` file is still blocked.
+- Operational reporting: when no logged-in gh account can read the pipeline repository, `usage-identity.mjs` reports under the active account (else the first), marked `usageLog.loginFallback` and cached for 1 day so an account with access added later takes over. A work machine often has only an enterprise-managed account logged in, which can never be a collaborator of a personal repository; the server decides whether it accepts that login. Only a machine with no gh account at all reports nothing.
+- `usage-register.mjs --quiet` prints one stderr line, `multi-agent: usage reporting is off - <reason>`, when reporting stays off for a reason the user can fix, and the Phase 0 backstop no longer discards stderr. A refused login is told to add the account with access with `gh auth login`.
+- The usage emitter waits 6 s for the endpoint instead of 2.5 s, so a cold start does not drop the record.
+- `smoke-cross-cli-behavior.sh` checks for any `$USER_`-prefixed login in the command docs instead of one hard-coded account suffix.
+
 ## [20.16.0] - 2026-10-02
 
 ### Added

package/README.md

@@ -158,7 +158,7 @@ The discipline behind all of this - bounded loops, evidence gates, token-budgete
 
 ## Commands
 
-`/multi-agent` plus 63 sub-commands. `/multi-agent:help` renders the same catalog in your terminal, in your `outputLanguage`.
+`/multi-agent` plus 64 sub-commands. `/multi-agent:help` renders the same catalog in your terminal, in your `outputLanguage`.
 
 ### Pipeline entries
 
@@ -219,6 +219,7 @@ The discipline behind all of this - bounded loops, evidence gates, token-budgete
 | `/multi-agent:test-accessibility`        | VoiceOver labels, sub-44pt tap targets, contrast, traits                  |
 | `/multi-agent:test-dynamic-type`         | Re-walk every screen at XL through accessibility-XL, report truncation    |
 | `/multi-agent:test-screenshots [locale]` | App Store screenshot set in a locale (defaults to `tr`)                   |
+| `/multi-agent:bug-bash`                  | Parallel exploratory sessions by charter; only findings a failing repro test proves are reported |
 | `/multi-agent:manual-test`               | Phase 3 standalone: check out the task branch and prepare it for Xcode    |
 
 ### Design, build and store
@@ -394,17 +395,17 @@ This enables the matching plugin (+ the shared `ai-common` plugin) in the repo's
 
 ## Tool support
 
-The pipeline runs natively on **Claude Code**, **Copilot CLI** and **Codex CLI** - all three install from the same `pipeline/` source and get the same 63 commands.
+The pipeline runs natively on **Claude Code**, **Copilot CLI** and **Codex CLI** - all three install from the same `pipeline/` source and get the same 64 commands.
 
 | Tool        | Flag                 | What it installs                                                                                       |
 | ----------- | -------------------- | ------------------------------------------------------------------------------------------------------ |
 | Claude Code | `--claude` (default) | slash commands + skills + agents + three `PreToolUse` hooks (secret scan, agent-guard, read-size gate) |
-| Copilot CLI | `--copilot`          | instructions + 63 sub-command skills + scripts                                                         |
-| Codex CLI   | `--codex`            | one router skill + 63 specs as refs + 10 agent TOML + `AGENTS.md` block + `codex mcp add`               |
+| Copilot CLI | `--copilot`          | instructions + 64 sub-command skills + scripts                                                         |
+| Codex CLI   | `--codex`            | one router skill + 64 specs as refs + 10 agent TOML + `AGENTS.md` block + `codex mcp add`               |
 
 Filter skills by stack with `--platform=ios\|android\|all`.
 
-**Why Codex gets one skill and not 63.** Codex assembles every discovered skill's name
+**Why Codex gets one skill and not 64.** Codex assembles every discovered skill's name
 and description into a single prompt block and drops entries when it overflows, with no
 error. Measured on 0.145: installing one plugin that declares 142 skills surfaced only
 75 of them and evicted an unrelated user skill. So on Codex the pipeline ships a single

package/README.tr.md

@@ -157,7 +157,7 @@ Koşunun kendisinde ayarlanabilen tek şey autopilot; geri kalan her şey kendi
 
 ## Komutlar
 
-`/multi-agent` ve 63 alt komut. `/multi-agent:help` aynı katalogu terminalde, `outputLanguage` ayarına göre gösterir.
+`/multi-agent` ve 64 alt komut. `/multi-agent:help` aynı katalogu terminalde, `outputLanguage` ayarına göre gösterir.
 
 ### Pipeline girişleri
 
@@ -218,6 +218,7 @@ Koşunun kendisinde ayarlanabilen tek şey autopilot; geri kalan her şey kendi
 | `/multi-agent:test-accessibility`     | VoiceOver etiketleri, 44pt altı dokunma alanları, kontrast, trait'ler           |
 | `/multi-agent:test-dynamic-type`      | Her ekranı XL'den accessibility-XL'e kadar yeniden gezer, kırpılmaları raporlar |
 | `/multi-agent:test-screenshots [dil]` | Bir dilde App Store screenshot seti (varsayılan `tr`)                           |
+| `/multi-agent:bug-bash`               | Charter başına paralel keşif oturumları; yalnız fail eden bir repro testinin kanıtladığı bulgular raporlanır |
 | `/multi-agent:manual-test`            | Faz 5'in bağımsız hali: görev branch'ine geçer, Xcode için hazırlar             |
 
 ### Tasarım, build ve store
@@ -393,17 +394,17 @@ Bu, ilgili plugin'i (+ ortak `ai-common` plugin'ini) repo'nun `.claude/settings.
 
 ## Araç desteği
 
-Pipeline **Claude Code**, **Copilot CLI** ve **Codex CLI** üzerinde native çalışır - üçü de aynı `pipeline/` kaynağından kurulur ve aynı 63 komutu alır.
+Pipeline **Claude Code**, **Copilot CLI** ve **Codex CLI** üzerinde native çalışır - üçü de aynı `pipeline/` kaynağından kurulur ve aynı 64 komutu alır.
 
 | Araç        | Bayrak                  | Ne kurar                                                                                                         |
 | ----------- | ----------------------- | ---------------------------------------------------------------------------------------------------------------- |
 | Claude Code | `--claude` (varsayılan) | slash komutları + skill'ler + agent'lar + üç `PreToolUse` hook'u (secret scan, agent-guard, okuma-boyutu geçidi) |
-| Copilot CLI | `--copilot`             | talimatlar + 63 alt-komut skill'i + script'ler                                                                   |
-| Codex CLI   | `--codex`               | bir router skill + ref olarak 63 spec + 10 agent TOML + `AGENTS.md` bloğu + `codex mcp add`                       |
+| Copilot CLI | `--copilot`             | talimatlar + 64 alt-komut skill'i + script'ler                                                                   |
+| Codex CLI   | `--codex`               | bir router skill + ref olarak 64 spec + 10 agent TOML + `AGENTS.md` bloğu + `codex mcp add`                       |
 
 Skill'leri stack'e göre filtrele: `--platform=ios\|android\|all`.
 
-**Codex neden 63 değil de tek bir skill alıyor.** Codex, keşfettiği her skill'in adını
+**Codex neden 64 değil de tek bir skill alıyor.** Codex, keşfettiği her skill'in adını
 ve açıklamasını tek bir prompt bloğuna toplar ve blok taştığında girdileri hatasızca
 düşürür. 0.145 üzerinde ölçüldü: 142 skill deklare eden bir plugin kurulduğunda sadece
 75'i yüzeye çıktı ve alakasız bir kullanıcı skill'i tahliye edildi. Bu yüzden Codex'te

package/docs/architecture.md

@@ -113,7 +113,7 @@ graph TB
     end
 
     subgraph "Pipeline Specs"
-        CMD[commands/<br/>63 command files]
+        CMD[commands/<br/>64 command files]
         AGT[agents/<br/>10 agent personas]
         RUL[rules/<br/>13 domain rules]
         PHS[multi-agent-refs/phases/<br/>phase specs + contracts]
@@ -204,8 +204,8 @@ repo and the website, plus the two independently shipped repos
 ```mermaid
 graph TD
     CC["Claude Code<br/>(source of truth)"]
-    COP["Copilot CLI<br/>(instructions + 63 sub-command skills)"]
-    COD["Codex CLI<br/>(1 router skill + 63 refs)"]
+    COP["Copilot CLI<br/>(instructions + 64 sub-command skills)"]
+    COD["Codex CLI<br/>(1 router skill + 64 refs)"]
     REPO["Pipeline Repo<br/>(npm package)"]
     WEB["Website"]
     PLUGREPO["multi-agent-plugins<br/>(6 plugins, own repo)"]

package/docs/ecosystem.md

@@ -5,7 +5,7 @@ separately, wired together at install time and at run time:
 
 | Repo | What it owns | Ships as |
 |---|---|---|
-| **`multi-agent-pipeline`** (this repo) | Orchestration: the 6-phase flow, the 63 slash commands, quality gates, review/triage, cross-CLI parity | npm package (`@mmerterden/multi-agent-pipeline`), installs itself onto Claude Code / Copilot CLI / Codex CLI |
+| **`multi-agent-pipeline`** (this repo) | Orchestration: the 6-phase flow, the 64 slash commands, quality gates, review/triage, cross-CLI parity | npm package (`@mmerterden/multi-agent-pipeline`), installs itself onto Claude Code / Copilot CLI / Codex CLI |
 | **`multi-agent-plugins`** | Stack knowledge: per-platform component/lifecycle skills (iOS, Android, Frontend, Backend) + shared knowledge | Claude Code marketplace, 6 independently-versioned plugins |
 | **`multi-agent-toolkit-mcp`** | The pipeline's hands on devices and browsers: 118 MCP tools across 14 categories (simulator/emulator control, memory, crash diagnostics, accessibility audit, store compliance, web automation, Figma-vs-mock design audit, code intelligence, wallet passes, an agent-DSL batch runner, a full-text context index, provider-backed research, video key frames, offline security scoring) | npm package, registered as a standard stdio MCP server on every host |
 
@@ -18,7 +18,7 @@ Either can be swapped or removed without touching the other two's source.
 graph LR
     subgraph PIPE ["multi-agent-pipeline (orchestrator)"]
         direction TB
-        PHASES["6 phases · 63 commands"]
+        PHASES["6 phases · 64 commands"]
         GATES["deterministic gates + review triage"]
     end
 
@@ -68,8 +68,8 @@ only those:
 graph TD
     CC["Claude Code<br/>~/.claude/commands/multi-agent/<br/>(source of truth)"]
 
-    CC -->|"Step 2: copy + reformat<br/>63 sub-command skills"| COP["Copilot CLI<br/>~/.copilot/skills/"]
-    CC -->|"Step 2b: transform<br/>(install.js --codex)"| COD["Codex CLI<br/>1 router skill + 63 refs<br/>+ 10 agent TOML"]
+    CC -->|"Step 2: copy + reformat<br/>64 sub-command skills"| COP["Copilot CLI<br/>~/.copilot/skills/"]
+    CC -->|"Step 2b: transform<br/>(install.js --codex)"| COD["Codex CLI<br/>1 router skill + 64 refs<br/>+ 10 agent TOML"]
     CC -->|"Step 3: genericize<br/>(strip personal data)"| REPO["multi-agent-pipeline repo<br/>pipeline/"]
     CC -->|"Step 4: version + feature sync"| WEB["Website<br/>projects.ts / i18n.tsx"]
 
@@ -161,7 +161,7 @@ measurements behind this table):
 
 | | Claude Code | Copilot CLI | Codex CLI |
 |---|---|---|---|
-| **Pipeline commands** | 63 slash-command skills, native | 63 sub-command skills, `multi-agent-{cmd}` naming, copied in | 1 router skill (`multi-agent`) + 63 command specs as reference files - Codex silently truncates its skills block past a few dozen entries, so sub-commands are not peer skills here |
+| **Pipeline commands** | 64 slash-command skills, native | 64 sub-command skills, `multi-agent-{cmd}` naming, copied in | 1 router skill (`multi-agent`) + 64 command specs as reference files - Codex silently truncates its skills block past a few dozen entries, so sub-commands are not peer skills here |
 | **Stack plugins** | Marketplace plugin, loaded natively, resolved by `.claude/settings.json` enabled-list | Enabled plugin's authored skills copied flat into `~/.copilot/skills/`; `knowledge/` **not** re-copied (already delivered via `shared/external`) | Copied as reference files under `~/.codex/multi-agent-refs/skills/`, plugin-prefixed on name clash (e.g. `architecture` → `ai-ios-toolkit-architecture`) |
 | **Component dispatch (Phase 3)** | Marketplace plugin's `create-component`/`create-screen` skill via the Skill tool | No plugin loader - the enabled stack plugin's authored skills (incl. `create-component`) are copied flat into `~/.copilot/skills/` at install time (the old frozen `figma-*` copies are pruned, they were never a fallback) | Not part of the enforced parity axis; classification + state-shape must match, skill *inventory* does not |
 | **multi-agent-toolkit-mcp** | `claude mcp add multi-agent-toolkit -- npx -y @mmerterden/multi-agent-toolkit-mcp` | `copilot mcp add multi-agent-toolkit -- npx -y @mmerterden/multi-agent-toolkit-mcp` | `codex mcp add multi-agent-toolkit -- npx -y @mmerterden/multi-agent-toolkit-mcp` (skipped with a warning if `codex` isn't on `PATH`) |

package/docs/facts.json

@@ -1,7 +1,7 @@
 {
   "$comment": "Generated by pipeline/scripts/gen-facts.mjs. Do not hand-edit: the site reads this, and a number edited here instead of at its source is the drift this file removes.",
-  "generatedAt": "2026-10-02",
-  "version": "20.16.0",
+  "generatedAt": "2026-10-07",
+  "version": "20.17.1",
   "phaseSchema": 2,
   "phases": [
     {
@@ -35,17 +35,17 @@
     "autopilot": [0, 1, 2, 3, 4, 5],
     "analysis": [0, 1, 3, 4, 5]
   },
-  "commandCount": 63,
-  "skillCount": 154,
-  "skillCountAll": 221,
+  "commandCount": 64,
+  "skillCount": 156,
+  "skillCountAll": 223,
   "agentCount": 10,
   "hosts": ["Claude Code", "Codex CLI", "Copilot CLI"],
-  "toolCount": 164,
+  "toolCount": 174,
   "toolCategories": {
-    "ios": 40,
-    "android": 31,
+    "ios": 42,
+    "android": 33,
     "store": 30,
-    "web": 18,
+    "web": 21,
     "vphone": 10,
     "code": 8,
     "design": 7,
@@ -53,10 +53,12 @@
     "pass": 4,
     "context": 3,
     "agent": 2,
+    "document": 2,
     "research": 2,
     "security": 2,
-    "media": 1
+    "media": 1,
+    "ui": 1
   },
-  "pluginSkillCount": 245,
-  "toolkitVersion": "3.18.0"
+  "pluginSkillCount": 247,
+  "toolkitVersion": "3.19.0"
 }

package/docs/features.md

@@ -161,7 +161,7 @@ When a specialized skill takes over a main pipeline phase, progress is reported
 
 ### Draft vs Ready Prompt
 
-Phase 4 asks `DRAFT or READY?` before creating the PR and persists the choice in `prefs.projects[].defaultPrMode`. An unattended run does not ask: the autopilot runner always opens a draft.
+Phase 4 asks `READY or DRAFT?` before creating the PR, recommends READY because every gate has passed by then, and persists the choice in `prefs.projects[].defaultPrMode`. An unattended run does not ask: the autopilot runner always opens a draft, and a person marks it ready.
 
 - Bitbucket: `draft: true` flag (DC 8.x+) with `[DRAFT]` title fallback for older servers.
 - GitHub: `gh pr create --draft` + `gh pr ready` for promotion.
@@ -355,7 +355,7 @@ Commands (`commands.mjs --json`) and run questions (`launch-request.mjs question
 - **Pre-Commit Secret Detection** (12 patterns): `PreToolUse` hook scans staged files for API keys/tokens, AWS access keys, private keys, `.env` files, service account JSON. Commit **blocked** if found.
 - **Read-Size Gate** (opt-in, `prefs.global.bulkRead.mode`): a `PreToolUse` hook inspects `Read` and the shell commands that read a file whole. In `observe` it only logs what it would have caught - the baseline you measure before routing anything. In `enforce` a file over `minLines` (default 350) is blocked and delegated to a haiku-rung worker (`bulk-read.sh`), which returns a line-numbered summary so the follow-up is a bounded `Read(offset:limit:)` instead of the whole file; the full text is parked under `.multi-agent/refs/`. The development phase and any file the run has already touched are exempt, because Claude Code's `Edit` requires its own `Read` first.
 - **Capture Hooks** (`SessionEnd`, `PreCompact`, `SessionStart`): every durable write used to live in Phase 5, the phase a run is least likely to reach. `SessionEnd` flushes a run that never got there; `PreCompact` flushes before an auto-compaction summarizes a long phase mid-flight, which is the same loss one level down; `SessionStart` prints at most two lines about an unfinished run. None calls a model, none reads a payload, and all exit 0 on every path - a hook that fails a session over bookkeeping is worse than the bookkeeping.
-- **Operational Reporting** (`prefs.global.usageLog`): coarse run metadata - task id and type, current phase, run status (in progress, completed, halted, parked), phase durations, whether a PR opened, token counts - sent at every phase transition and at completion, halt and park, and never prompts, code, diffs or absolute paths (task title and PR URL only when `includeTitle` / `includePrUrl` are set). Runs are reported under the logged-in GitHub account that can read the pipeline repository (`usage-identity.mjs`, cached in `usageLog.login`, `--refresh` to re-check); with no such account nothing is sent. The per-machine token is REQUESTED from the endpoint by `usage-register.mjs` (setup, update, and the Phase 0 exit gate as a backstop), is write-only, and lives in the OS credential store; prefs hold only the entry name and the switch. `usageLog.optOut: true` blocks registration permanently and is checked before the network call. The server issues a token only to the repository owner or a collaborator, pins the login to it, and shows only those accounts' records on the panel. An unreachable endpoint leaves reporting off with one line and exit 0 - a run is never failed over bookkeeping.
+- **Operational Reporting** (`prefs.global.usageLog`): coarse run metadata - task id and type, current phase, run status (in progress, completed, halted, parked), phase durations, whether a PR opened, token counts - sent at every phase transition and at completion, halt and park, and never prompts, code, diffs or absolute paths (task title and PR URL only when `includeTitle` / `includePrUrl` are set). Runs are reported under the GitHub account on the machine that can read the pipeline repository, found with no question asked through gh, the git credential helper or an SSH key (`usage-identity.mjs`, cached in `usageLog.login` / `loginMethod`, `--refresh` to re-check); with no such account nothing is sent. The per-machine token is REQUESTED from the endpoint by `usage-register.mjs` (setup, update, and the Phase 0 exit gate as a backstop), is write-only, and lives in the OS credential store; prefs hold only the entry name and the switch. `usageLog.optOut: true` blocks registration permanently and is checked before the network call. Registration proves the login with that account's gh or credential-helper token, or with an SSH signature over a server challenge. The server issues a token only to the repository owner or a collaborator, pins the login to it, and shows only those accounts' records on the panel. An unreachable endpoint leaves reporting off with one line and exit 0 - a run is never failed over bookkeeping.
 - **Build Queue**: All `xcodebuild` calls acquire a lock. Each worktree uses own `-derivedDataPath`. Stale locks auto-clean after 15 min. Non-Xcode builds don't need the lock.
 - **Context Management**: `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=65` - compaction at 65% usage (prevents degradation in 6-phase sessions).
 - **3-Iteration Hard Kill**: Any retry loop stops after 3 attempts, then pauses for user. No infinite loops.
@@ -382,6 +382,14 @@ Adversarial fixtures that test triage resilience against adversarial reviewer ou
 
 Detects drift between Claude Code instructions (`~/.claude/commands/multi-agent/SKILL.md`), Copilot CLI instructions (`~/.copilot/copilot-instructions.md`), and the repo's pipeline spec files. Reports discrepancies during Phase 0 Init.
 
+### Exploratory Testing and Bug Bash
+
+`/multi-agent:test` runs every exploratory session under one contract (`features/exploratory-testing.md`, `explore-findings.mjs`, `explore-run.schema.json`): each finding has expected, observed, repro steps, severity 1-5, kind `issue` or `warning`, and evidence on disk; each run has a step, minute and token budget and stops after three consecutive blocked steps that reported nothing; steps blocked by environment, credentials, seed data or the automation itself are reported apart from product failures. Credentials are typed by `secret_ref`, and locators plus expect steps replace coordinates when the toolkit offers them.
+
+A visual or semantic assertion is decided by a fresh judge subagent that receives only the assertion and the current screenshot or UI tree (`features/assertion-judge.md`); `explore-findings.mjs judge-packet` refuses any field describing the run, and an inconclusive or uncited answer fails as `ASSERTION_INCONCLUSIVE`.
+
+`/multi-agent:bug-bash` fans out 5-10 charters, each with a persona and its own budget, merges the findings, triages them against the source, and proves every candidate with a reproducing test. `bug-bash.mjs report` confirms a finding only when its repro failed with `ASSERTION_FAILED` on the assertion that encodes it; everything else is listed as rejected with the reason (artifact, environment, design, fixture, did-not-reproduce, unproven, unverified). The `e2e-testing` knowledge skill in ai-common-toolkit carries the toolkit side: locators, expect verdicts, traces, replay, parallel web sessions and synthetic MRZ document fixtures.
+
 ### Token Budget Enforcement
 
 Per-phase token budgets prevent runaway sessions. If a phase exceeds its budget, the pipeline pauses and offers: continue (extend budget), skip phase, or abort. Budgets are configurable in `prefs.global.tokenBudgets`.