create-byan-agent 2.23.0 → 2.26.0

package/CHANGELOG.md

@@ -9,6 +9,236 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.26.0] - 2026-06-16
+
+### Added - Leantime FD auto-sync hook (FD lifecycle -> board, automatic)
+
+The FD -> Leantime mirror is now AUTOMATIC. A `PostToolUse` hook
+(`.claude/hooks/leantime-fd-sync.js`, registered in `.claude/settings.json`) fires
+after `byan_fd_advance` / `byan_fd_update` and drives the board with no agent
+action: it ensures the project at DISCOVERY, creates one task per backlog feature
+at DISPATCH, and moves tasks through `todo -> doing -> blocked/review -> done` as
+the FD advances. This supersedes the hand-driven section 2.5 fire points (which
+the agent had to run by hand and could skip).
+
+- **Pure core** (`_byan/mcp/byan-mcp-server/lib/leantime-fd-core.js`):
+  `decideActions` maps a phase transition + the sidecar to ordered Leantime
+  intents; unit-tested for every transition. The hook is a thin I/O shell that
+  executes them.
+- **Best-effort + bounded**: the hook exits 0 in every path (a sync issue does
+  not block the turn), no-ops when Leantime is off, self-heals a dropped call on
+  the next phase event (a per-call timeout + a hook wall-clock budget), and logs
+  every attempt to `.byan-leantime/sync.jsonl`.
+- **Idempotence**: a gitignored sidecar (`.byan-leantime/map.json`, keyed by
+  fd_id) is the single id ledger — a REFACTOR loop re-builds without duplicating a
+  project or task. The hook does not write `fd-state.json` (state-coupling).
+- **Human visibility** (`assignUserToProject` + `LEANTIME_ASSIGN_USER_ID`): an
+  API-created project is owned by the API service user and hidden from a person's
+  project selector; the hook relates the configured human so the board shows up.
+  The underlying Leantime RPC reconciles a user's whole project list, so the
+  assign reads the full list first and writes the union (fail-closed if that read
+  is incomplete) to avoid unassigning the user's other projects.
+
+### Added - Leantime project-management integration (one-way FD -> board)
+
+BYAN can now mirror its Feature Development lifecycle onto a self-hosted Leantime
+instance. When `LEANTIME_API_URL` + `LEANTIME_API_TOKEN` are configured, the FD
+phases drive a Leantime project and one task per backlog feature ; when absent,
+the tools report disabled and FD proceeds unchanged. The sync is one direction
+(FD -> Leantime) and best-effort : a down or misconfigured Leantime degrades to
+`{ synced:false, reason }` and does not block a phase transition.
+
+- **Client** (`_byan/mcp/byan-mcp-server/lib/leantime-sync.js`): a JSON-RPC 2.0
+  client for `<base>/api/jsonrpc`, authenticated by the Leantime-native
+  `x-api-key` header (kept distinct from the byan_web `ApiKey/Bearer` scheme).
+  Best-effort and does not throw, with an `AbortController` timeout and a
+  non-JSON-200 guard that rejects an HTML login body (the wrong-host lesson)
+  instead of reading it as an empty board. Business fns: `ensureProject`
+  (idempotent by name), `createTask`, `moveTask`, `assignTask`, `getTask`,
+  `getBoard`, plus `resolveStatusMap` / `resolveClientId` / `resolveEditorId`.
+- **MCP tools** (7): `byan_leantime_ping`, `byan_leantime_project_ensure`,
+  `byan_leantime_task_create`, `byan_leantime_task_move`,
+  `byan_leantime_task_assign`, `byan_leantime_task_get`,
+  `byan_leantime_board_get`. All but `ping` pass through `requireLeantime()`.
+- **FD wiring** (`.claude/skills/byan-byan/SKILL.md` section 2.5): fire points
+  DISCOVERY -> project_ensure, DISPATCH -> task_create per feature, BUILD ->
+  doing, REVIEW/VALIDATE-KO -> blocked, VALIDATE-OK -> review, DOC -> done.
+  Leantime ids persist into fd-state (`project_context.leantime.projectId`,
+  backlog `leantime.taskId`) so a REFACTOR loop reuses tasks instead of
+  duplicating them.
+- **Status mapping**: the canonical FD columns (`todo|doing|blocked|review|done`)
+  resolve to per-project Leantime status ids at call time, with a conservative
+  fallback when the labels cannot be read.
+- **Tests**: `test/leantime-sync.test.js` (14 cases: auth header, non-JSON guard,
+  timeout, idempotence, column resolution) + `test/leantime-tools.test.js`
+  (the 7-tool declaration/handler surface in server.js).
+- **Docs**: `.claude/rules/byan-api.md` section 8 (the `byan_leantime_*` family +
+  the wrong-host lesson).
+- **Pending**: the live wire-format verification (one real POST with a Leantime
+  PAT, to confirm `params:{values:{}}` wrapping against the running instance) is
+  a documented follow-up ; the tested format follows the Leantime master source.
+
+### Removed - GitHub Copilot CLI + VSCode dropped as target platforms (3 -> 2)
+
+BYAN now targets two platforms: Claude Code and Codex. GitHub Copilot CLI and
+the VSCode extension are no longer install targets. This is a breaking change for
+anyone who relied on the Copilot/VSCode output.
+
+- **Install path** (`npx create-byan-agent`): the platform menu, auto-select, and
+  generated stubs cover Claude Code + Codex only. The `byan_copilot_*` MCP tool
+  family and the `marc` Copilot-oriented agent were removed in the core pass
+  (commit `0f06cf8`).
+- **Web UI** (ships on npm via `install/src/webui`): the chat CLI selector, the
+  `cli-detector` definitions, the platform-detection list, and the marketing copy
+  drop Copilot and VSCode; the `copilot-adapter` bridge and its `createBridge`
+  case are gone (`createBridge('copilot')` now rejects with `Unknown CLI adapter`).
+- **Dead code**: both shipped copies of the orphaned `copilot-context` module
+  (`src/byan-v2/context/` and `install/src/byan-v2/context/`) and two stale
+  non-jest harnesses (`test-byan-v2-workflow.js`, `test-workflow-simple.js`) were
+  deleted.
+- **Note**: the `byan-loadbalancer` Copilot *provider* (an LLM backend, not an
+  install target) is unaffected and stays.
+
+### Added - Auto-Benchmark: native sourced decision benchmarks (C1-C5)
+
+When the agent is about to ask you to choose between options, it now benchmarks
+the fork by default: one compact `Option | criteria | Niv` table with a best-first
+recommendation, sourced and confidence-tagged, at the right level of detail — so
+you no longer have to ask each time. Two layers cover this honestly (Claude Code
+exposes no pre-display interception hook today, GH #28273):
+
+- **Proactive doctrine** (the broadly-portable layer). The full doctrine lives in
+  `.claude/rules/benchmark.md`, generated from the single source of truth
+  `_byan/_config/autobench.yaml` by `byan-sync-rules`, and a lean pointer is
+  upserted cross-platform into `.claude/CLAUDE.md`, `AGENTS.md`, and
+  `.github/copilot-instructions.md` (idempotent `BYAN-AUTOBENCH` markers). It
+  covers the TRIGGER 2-gate rule (>= 2 non-substitutable options diverging on
+  >= 1 weighted criterion) + the exemption list (y/n confirms, destructive
+  prompts) + internal/external routing + a verbatim few-shot decision tree, the
+  SCALER 5-level evidence rubric + strict-domain floors + the link-only-if-WebFetch
+  rule, the FORMAT compact table with hard caps (<= 4 options / <= 4 criteria /
+  <= 3 links) + collapse-the-degenerate + `[bench:expand]` opt-in, and the
+  ANTI-BLOAT latency guard + escape-hatch + no-re-benchmark.
+- **Reactive Stop hook** (`.claude/hooks/autobench-stop-guard.js`), the safety
+  net. It **ships DISARMED**: it observes and ledgers every turn but stays inert
+  (does not block) until you opt in — set `enforcement.armed: true` in
+  `_byan/_config/autobench.yaml` and run `byan-sync-rules` (config-only; there is
+  no loose flag file) — so day one is zero noise / zero latency.
+  Detection is **artifact-primary**: a real fork is recognized from an
+  `AskUserQuestion` tool_use in the finished turn, with the choice-language regex
+  as a last-resort fallback. Block-once is content-hashed (no loop); a session
+  escape-hatch (`touch .byan-autobench/off`) plus a cross-session toggle suppress
+  it.
+- **Tooling.** A `byan-benchmark` skill (conductor) and a DATA-only native
+  workflow (`.claude/workflows/byan-benchmark.js`), both registered in the
+  workflow manifest / PORTABLE bucket / INDEX. A BYAN-only opt-in layer enriches
+  the matrix via `byan_fc_check`. Every fire/miss is audited to
+  `_byan-output/benchmark-ledger.jsonl`.
+
+### Added - byan-install-core (F1): headless, deterministic install engine
+
+First feature of the installer refactor (FD lot 1). A new internal workspace
+package `byan-install-core` (`install/packages/install-core/`) that replaces the
+LLM-driven AUTO interview with a deterministic engine. It is the shared core both
+front-ends bind to: the npm CLI wizard (F2, next) and the Electron app (F5).
+
+- **Four-verb lifecycle.** `detect(opts)` builds a serializable MachineProfile
+  (os, arch, node, npm, git, claude, codex) and is spawn-free by default
+  (`probeVersions` is opt-in). `plan(profile, answers)` is pure and turns the
+  non-interactive answers contract into an ordered InstallPlan. `apply(plan, opts)`
+  is the only mutator. `verify(plan|target)` is read-only.
+- **No LLM, no `which`.** Detection uses a pure-Node PATH walk (`lookpath`,
+  honoring Windows PATHEXT); recommendations come from a versioned JSON decision
+  table (`data/recommender.json`), not a model call.
+- **Per-OS env + validate-or-die .mcp.json.** `env-writer` and `mcp-renderer`
+  reuse the existing `byan-platform-config` package (idempotent marker blocks;
+  a config is parsed and validated before any write, so a broken file is not
+  emitted on an invalid input).
+- **Per-user install, no sudo.** `--install-cli claude|codex` installs via
+  `npm -g`; the AUTH step is an explicit handoff (the engine returns the manual
+  command and reports a pending state rather than a fake authenticated success).
+- **ES5 preflight.** `byan-install-core/preflight` is a dependency-free, ES5-only
+  entry a launcher can require on an ancient Node to gate the version before any
+  modern module loads.
+- Tests: 9 suites / 161 tests for the package; full repo suite green
+  (no regression). The CLI wizard, `doctor`, journal/resume and the v2.19
+  `--yes` end-to-end snapshot land in the following features (F2, F3).
+
+## [2.25.0] - 2026-06-09
+
+### Added - Advisory auto-feed (BYAN learns from each session, automatically)
+
+The insight loop observed and proposed; the missing half was the LEARNING. BYAN's
+advisory ledgers (ELO trust, the suitability ledger) updated only when the agent
+remembered to call a record tool. This wires the automatic half — outcomes are
+recorded at end of turn, with no agent action — while behavior surfaces stay
+human-gated.
+
+- **Capture.** The `byan_outcome_log` MCP tool appends one validated advisory
+  outcome to a buffer (cheap; it does not write a ledger directly). kind=elo logs
+  `{domain, result}`; kind=suitability logs `{model, leafId, success}`.
+- **Drain.** `.claude/hooks/drain-advisory.js` is a Stop hook that, at end of each
+  turn, records the buffered outcomes into the ELO ledger (full Glicko update) and
+  the suitability ledger, advancing a line cursor for idempotency. It is strictly
+  non-blocking (all work in try/catch, emits `{continue:true}` and exit 0 on every
+  path) and crosses the ESM/CJS boundary (the CJS ELO engine via require, the ESM
+  suitability store via dynamic import).
+- **Advisory-only.** The loop writes only the buffer and the two advisory ledgers.
+  Behavior surfaces (routing, personas, mantra thresholds) are left untouched —
+  those stay a human decision, consistent with the insight loop's gated philosophy.
+- 71 tests (the pure planners, the buffer, and a drain-hook e2e with ledger
+  snapshot/restore) plus a live smoke test recording a real Glicko update. The tool
+  and hook ship in the template; the hook registers alongside the existing Stop
+  hooks.
+- Explicit follow-ups (out of this scope): the adversarial verdict panel that would
+  feed suitability without a manual log, and a fact-graph-derived ELO source.
+
+## [2.24.0] - 2026-06-09
+
+### Added - Session insight loop (gated self-improvement)
+
+BYAN already has advisory learning surfaces (ELO trust, the suitability ledger)
+and the native Claude Code hooks already leave outcome trails on disk, but the
+loop was open: the agent had to read and act on them by hand. This closes it,
+under a strict gated philosophy.
+
+- **Harvester** `_byan/mcp/byan-mcp-server/lib/insight-harvest.js` +
+  `bin/byan-insight-digest.js` + the `byan_insight_digest` MCP tool: read the
+  native trails (`tool-log.jsonl` health, strict `audit.log` recurring gaps,
+  the suitability ledger routing outcomes, the ELO profile trends) and aggregate
+  them into a digest with conservative, GATED proposals. Pure aggregation +
+  IO-isolated reader, mirroring the template-fidelity pattern.
+- **Gated by design.** The harvester only READS; it writes nothing to a behavior
+  surface (routing, personas, mantra thresholds). Every proposal carries
+  `gated: true` and is surfaced for a human to ratify — an agent that rewrote its
+  own routing on a heuristic would be the silent-downgrade BYAN exists to prevent.
+- **Skill** `byan-insight` presents the digest as a gated improvement proposal
+  (observe, propose, human ratifies), consistent with the advisory ELO /
+  suitability doctrine.
+- **Guard false-positive fix.** `tool-failure-guard` flagged any tool whose result
+  echoed the literal phrase "internal error" as a failure, exempting only
+  Write/Edit/Read. Bash (diagnostic stdout) and MCP tools (echoed stored data) now
+  join the echo-heavy set: their `is_error` flag is trusted, content patterns are
+  not. A genuine failure still sets `is_error`. Caught live (a Bash log-grep
+  blocked the session twice) and covered by unit + e2e tests.
+- 43 harvester unit tests + the detector tests; the e2e guard tests moved their
+  content-pattern cases onto a non-echo tool. The tool and skill ship in the
+  template.
+
+### Changed - Closed the fused-route and output-folder legacy debts
+
+- Removed the dead parallel router `src/core/dispatcher/execution-router.js` (zero
+  live consumers) and its test; the routing docs (`workers.md`,
+  `feature-workflow.md`) and the loadbalancer architecture comment now point only
+  to `byan_dispatch` and its two-axis model (strategy from score, model tier from
+  nature).
+- Standardized the documented output folder from the legacy `_bmad-output/` to the
+  runtime's `_byan-output/` across the agent and platform docs plus an inert config
+  default. Left untouched on purpose: the deliberate back-compat read in
+  `agent-packager.js` (recovers agent creations from older installs under
+  `_bmad-output/bmb-creations`), the migration guides, and the anti-regression
+  tests that assert the old name is gone.
+
 ## [2.23.0] - 2026-06-09
 
 ### Added - Stub path normalizer + a 5th pre-commit gate (no _bmad/@bmad drift)

package/README.md

@@ -69,7 +69,7 @@ Diversifier les personas elargit la surface de competence collective.
 
 - Node.js >= 12.0.0
 - npm >= 6.0.0
-- Un compte GitHub Copilot, Claude Code ou Codex (selon la plateforme cible)
+- Un compte Claude Code ou Codex (selon la plateforme cible)
 
 ### Installation rapide (recommandée)
 
@@ -89,7 +89,7 @@ L'installeur (Yanstaller) vous guide interactivement à travers le processus :
 ```
 ? Nom du projet : mon-projet
 ? Langue de communication : Francais
-? Plateforme cible : GitHub Copilot CLI
+? Plateforme cible : Claude Code
 ? Activer le fact-check scientifique ? [Y/n]
 ? Activer le système ELO de confiance ? [Y/n]
 ? Optimiser les coûts LLM automatiquement (~54% d'économies) ? [Y/n]
@@ -108,7 +108,6 @@ votre-projet/
     workflows/         # Workflows guidés
     knowledge/         # Base de connaissances sources (axiomes, benchmarks)
     config.yaml        # Configuration principale (langue, utilisateur, chemins)
-  .github/agents/      # Wrappers GitHub Copilot CLI
   .claude/             # Intégration Claude Code (si activée)
   .codex/              # Intégration Codex/OpenCode (si activée)
   bin/byan-v2-cli.js   # CLI BYAN
@@ -164,7 +163,7 @@ const ctx = byan.getClaimContext('security');
 console.log('Scaffold level :', ctx.scaffoldLevel);
 ```
 
-### Activer les agents dans GitHub Copilot CLI
+### Activer les agents dans Claude Code ou Codex
 
 Une fois installé, les agents sont disponibles via `@agent-name` :
 
@@ -183,7 +182,7 @@ Une fois installé, les agents sont disponibles via `@agent-name` :
 
 ## Liste des Agents Disponibles
 
-BYAN contient **27 agents spécialisés** organisés en **5 modules** :
+BYAN contient **26 agents spécialisés** organisés en **5 modules** :
 
 ### Core — Fondation de la Plateforme
 
@@ -203,7 +202,6 @@ BYAN contient **27 agents spécialisés** organisés en **5 modules** :
 | **agent-builder** | Bond | Expert en construction d'agents BMAD-compliant | Construire un agent complexe manuellement |
 | **module-builder** | Morgan | Architecte de modules BYAN complets | Créer un nouveau module métier |
 | **workflow-builder** | Wendy | Designer de workflows guidés | Concevoir un processus multi-étapes |
-| **marc** | Spécialiste | Intégration GitHub Copilot CLI | Déployer des agents sur Copilot |
 | **rachid** | Spécialiste | Déploiement npm/npx | Publier un package BYAN |
 | **carmack** | Optimiseur | Optimisation tokens (-46%) | Réduire le coût d'utilisation des agents |
 | **patnote** | Gestionnaire | Mises à jour BYAN et résolution de conflits | Mettre à jour un projet BYAN existant |
@@ -274,7 +272,7 @@ Un agent est un spécialiste IA avec une identité définie. Il possède :
 - **Rules** : les contraintes absolues qu'il ne peut pas violer (les 71 mantras)
 - **Capabilities** : ce qu'il peut faire, ce qu'il ne fait pas
 
-Les agents sont définis en Markdown avec des sections XML. Ils sont stockés dans `_byan/{module}/agents/` et exposés sur chaque plateforme via un wrapper léger (`.github/agents/`, `.claude/`, `.codex/prompts/`).
+Les agents sont définis en Markdown avec des sections XML. Ils sont stockés dans `_byan/{module}/agents/` et exposés sur chaque plateforme via un wrapper léger (`.claude/`, `.codex/prompts/`).
 
 ### Workflow
 
@@ -432,7 +430,7 @@ Domaines stricts : `security` / `performance` / `compliance` → LEVEL-2 minimum
 
 Mode d'enforcement qui empêche l'agent de livrer moins que demandé (un MVP au
 lieu de l'app prod, un stub au lieu de la feature, un template baclé). Actif sur
-les **3 plateformes** : Claude Code, Codex, GitHub Copilot.
+les **2 plateformes** : Claude Code, Codex.
 
 ```
 1. Lock du scope       byan_strict_lock_scope  (scope verbatim + critères testables)
@@ -442,8 +440,8 @@ les **3 plateformes** : Claude Code, Codex, GitHub Copilot.
 ```
 
 Le commit est **bloqué** par un filet pre-commit tant que la session strict
-engagée n'est pas complétée correctement — y compris pour Codex et Copilot qui
-n'ont pas de hook in-session. Source de vérité unique :
+engagée n'est pas complétée correctement — y compris pour Codex qui
+n'a pas de hook in-session. Source de vérité unique :
 `_byan/_config/strict-mode.yaml`, régénérée via `byan-sync-rules`.
 
 Activation : `byan_fd_start strict:true`, skill `byan-strict`, ou auto-détection
@@ -466,7 +464,7 @@ sur mots-clés (`prod`, `client`, `livrable`, `contrat`, `release`...).
 | `testarch-atdd` | Générer des tests ATDD avant implémentation | tea |
 | `fact-check` | Analyser une assertion ou un document | fact-checker |
 | `elo-workflow` | Consulter et gérer le score de confiance ELO | byan |
-| `byan-sync-rules` | Régénérer les artefacts du mode strict (3 plateformes) | byan |
+| `byan-sync-rules` | Régénérer les artefacts du mode strict (2 plateformes) | byan |
 
 ---
 
@@ -474,7 +472,6 @@ sur mots-clés (`prod`, `client`, `livrable`, `contrat`, `release`...).
 
 | Plateforme | Invocation | Chemin de config |
 |------------|-----------|------------------|
-| GitHub Copilot CLI | `@agent-name` | `.github/agents/*.md` |
 | Claude Code | `@agent-name` | `.claude/rules/*.md` |
 | Codex / OpenCode | `@agent-name` | `.codex/prompts/*.md` |
 | CLI direct | `node bin/byan-v2-cli.js` | `_byan/config.yaml` |