create-byan-agent 2.23.0 → 2.26.0

package/install/templates/_byan/mcp/byan-mcp-server/server.js

@@ -9,8 +9,10 @@ import {
   ListToolsRequestSchema,
 } from '@modelcontextprotocol/sdk/types.js';
 import { dispatch } from './lib/dispatch.js';
+import { harvest as harvestInsights, renderDigest as renderInsightDigest } from './lib/insight-harvest.js';
+import { appendOutcome } from './lib/outcome-buffer.js';
+import { validateForLog, eloOutcomeForStrictComplete } from './lib/advisory-autofeed.js';
 import { readSoul, appendSoulMemory } from './lib/soul.js';
-import { listSessions, readSessionEvents, searchSessions } from './lib/copilot.js';
 import {
   start as fdStart,
   status as fdStatus,
@@ -69,6 +71,17 @@ import {
   syncEnabled as strictSyncEnabled,
   resolveProjectId as strictResolveProjectId,
 } from './lib/strict-sync.js';
+import {
+  syncEnabled as leantimeEnabled,
+  rpc as leantimeRpc,
+  ensureProject as leantimeEnsureProject,
+  createTask as leantimeCreateTask,
+  moveTask as leantimeMoveTask,
+  assignTask as leantimeAssignTask,
+  getTask as leantimeGetTask,
+  getBoard as leantimeGetBoard,
+  METHODS as LEANTIME_METHODS,
+} from './lib/leantime-sync.js';
 
 // Compact view of a best-effort strict-sync result for tool responses.
 function syncResult(sync) {
@@ -110,6 +123,14 @@ function requireToken() {
   }
 }
 
+// Leantime uses its OWN env pair (LEANTIME_API_URL/LEANTIME_API_TOKEN), kept
+// distinct from BYAN_API_URL so the two backends never get crossed.
+function requireLeantime() {
+  if (!process.env.LEANTIME_API_URL || !process.env.LEANTIME_API_TOKEN) {
+    throw new Error('LEANTIME_API_URL + LEANTIME_API_TOKEN env vars are required for byan_leantime_* tools.');
+  }
+}
+
 async function apiRequest(path, options = {}) {
   const url = `${BYAN_API_URL}${path}`;
   const headers = {
@@ -404,39 +425,6 @@ const tools = [
       additionalProperties: false,
     },
   },
-  {
-    name: 'byan_copilot_sessions',
-    description:
-      'List GitHub Copilot CLI sessions stored locally at ~/.copilot/session-state/. Returns sessionId, start/end time, cwd, branch, agent name, message and tool call counts. Sorted most-recent-first. Use to discover past Copilot CLI conversations for reference or import.',
-    inputSchema: {
-      type: 'object',
-      properties: {
-        limit: { type: 'number', description: 'Max sessions to return (default 20).' },
-        sinceIso: { type: 'string', description: 'ISO timestamp filter — only sessions started after this.' },
-        cwdFilter: { type: 'string', description: 'Substring match on session cwd (e.g. "byan_web").' },
-      },
-      additionalProperties: false,
-    },
-  },
-  {
-    name: 'byan_copilot_session_events',
-    description:
-      'Read events of a specific Copilot CLI session (events.jsonl). Optionally filter by event type (user.message, assistant.message, tool.execution_start, etc.). Useful to inspect the flow of a past session.',
-    inputSchema: {
-      type: 'object',
-      properties: {
-        sessionId: { type: 'string', description: 'Session UUID from byan_copilot_sessions.' },
-        types: {
-          type: 'array',
-          items: { type: 'string' },
-          description: 'Filter to these event types only.',
-        },
-        limit: { type: 'number', description: 'Max events (default 200).' },
-      },
-      required: ['sessionId'],
-      additionalProperties: false,
-    },
-  },
   {
     name: 'byan_fd_start',
     description:
@@ -545,6 +533,34 @@ const tools = [
       additionalProperties: false,
     },
   },
+  {
+    name: 'byan_insight_digest',
+    description:
+      'Harvest native Claude Code outcome trails (tool-log, strict-audit gaps, the suitability ledger, ELO) into a GATED improvement digest for BYAN. Read-only: it OBSERVES and PROPOSES; every proposal is gated for a human to ratify, nothing is auto-applied to routing / personas / mantras. Returns { toolHealth, recurringGaps, routingOutcomes, eloTrends, proposals }.',
+    inputSchema: {
+      type: 'object',
+      properties: {},
+      additionalProperties: false,
+    },
+  },
+  {
+    name: 'byan_outcome_log',
+    description:
+      'Log one ADVISORY outcome to the auto-feed buffer (cheap append; it never writes a ledger directly). The drain-advisory Stop hook records buffered outcomes into the ELO / suitability ledgers at end of turn, so BYAN auto-learns without the agent recording by hand. kind=elo needs { domain, result: VALIDATED|PARTIAL|BLOCKED }; kind=suitability needs { model, leafId, success }. Advisory-only: behavior surfaces (routing / personas / mantras) are never written.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        kind: { type: 'string', enum: ['elo', 'suitability'] },
+        domain: { type: 'string', description: 'elo: the technical domain of the claim' },
+        result: { type: 'string', enum: ['VALIDATED', 'PARTIAL', 'BLOCKED'], description: 'elo: the claim verdict' },
+        model: { type: 'string', description: 'suitability: the cheap model tier/id' },
+        leafId: { type: 'string', description: 'suitability: the workflow leaf' },
+        success: { type: 'boolean', description: 'suitability: did the cheap model survive adversarial review' },
+      },
+      required: ['kind'],
+      additionalProperties: false,
+    },
+  },
   {
     name: 'byan_strict_lock_scope',
     description:
@@ -566,6 +582,10 @@ const tools = [
           items: { type: 'string' },
           description: 'Glob patterns of paths the agent may modify.',
         },
+        domain: {
+          type: 'string',
+          description: 'Optional explicit ELO domain (e.g. security, performance, javascript). When set, a successful byan_strict_complete feeds one VALIDATED outcome to the ELO learning loop. Recorded verbatim (your explicit input, never inferred from text); omit to feed nothing.',
+        },
         force: { type: 'boolean', description: 'Relock with different scope.' },
         projectId: {
           type: 'string',
@@ -627,7 +647,7 @@ const tools = [
   {
     name: 'byan_strict_suggest',
     description:
-      'Check whether a piece of text (user request, feature name) signals a production-grade deliverable that should be built under strict mode. Reads activation keywords from _byan/_config/strict-mode.yaml. Returns { suggested, matched, message }. Use on any platform (Codex/Copilot have no in-session hook) to decide whether to lock strict mode.',
+      'Check whether a piece of text (user request, feature name) signals a production-grade deliverable that should be built under strict mode. Reads activation keywords from _byan/_config/strict-mode.yaml. Returns { suggested, matched, message }. Use on any platform (Codex has no in-session hook) to decide whether to lock strict mode.',
     inputSchema: {
       type: 'object',
       properties: {
@@ -817,25 +837,6 @@ const tools = [
       additionalProperties: false,
     },
   },
-  {
-    name: 'byan_copilot_search',
-    description:
-      'Full-text search across all Copilot CLI sessions. Finds messages (user + assistant by default) containing the query string. Returns sessionId + timestamp + excerpt. Use to recall past discussions without knowing which session they were in.',
-    inputSchema: {
-      type: 'object',
-      properties: {
-        query: { type: 'string', description: 'Substring to search for (case-insensitive).' },
-        types: {
-          type: 'array',
-          items: { type: 'string' },
-          description: 'Event types to scan (default: user.message, assistant.message).',
-        },
-        limit: { type: 'number', description: 'Max matches (default 50).' },
-      },
-      required: ['query'],
-      additionalProperties: false,
-    },
-  },
 
   // ─── Projects ─────────────────────────────────────────────────────────
   {
@@ -1167,6 +1168,101 @@ const tools = [
       additionalProperties: false,
     },
   },
+
+  // ─── Leantime (project-management mirror) ─────────────────────────────
+  // Client-side automation of the self-hosted Leantime JSON-RPC API. Used by
+  // the FD workflow to create a project + a task per feature and move task
+  // status across phases. Needs LEANTIME_API_URL + LEANTIME_API_TOKEN.
+  {
+    name: 'byan_leantime_ping',
+    description:
+      'Healthcheck the Leantime integration: reports api_url, token presence, and (if configured) whether the JSON-RPC API is reachable. Surfaces the wrong-host guard (HTML instead of JSON). No required args.',
+    inputSchema: { type: 'object', properties: {}, additionalProperties: false },
+  },
+  {
+    name: 'byan_leantime_project_ensure',
+    description:
+      'Idempotent create-or-fetch of a Leantime project from the FD project_context. Matches an existing project by name first (no duplicate on FD re-run). Returns { id, created }. Requires LEANTIME_API_*.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        name: { type: 'string', description: 'Project name (defaults to slug).' },
+        slug: { type: 'string', description: 'Project slug (fallback name).' },
+        clientId: { type: 'number', description: 'Owning Leantime client id. Resolved if omitted.' },
+        details: { type: 'string', description: 'Optional project description.' },
+      },
+      additionalProperties: false,
+    },
+  },
+  {
+    name: 'byan_leantime_task_create',
+    description:
+      'Create one Leantime task (ticket) from an FD backlog item. Returns the new task id to store back in fd-state (caller owns idempotency: create only if the item has no leantime_task_id). Requires LEANTIME_API_*.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        projectId: { type: 'number', description: 'Leantime project id.' },
+        headline: { type: 'string', description: 'Task title.' },
+        description: { type: 'string' },
+        status: { type: 'number', description: 'Leantime status id (optional).' },
+        priority: { type: 'number' },
+        editorId: { type: 'number', description: 'Assignee/editor user id.' },
+        tags: { type: 'string' },
+        type: { type: 'string', description: "Ticket type, default 'task'." },
+      },
+      required: ['projectId', 'headline'],
+      additionalProperties: false,
+    },
+  },
+  {
+    name: 'byan_leantime_task_move',
+    description:
+      'Move a Leantime task to a lifecycle column (todo|doing|blocked|review|done). Resolves the column to the project status id, then updates the ticket. Requires LEANTIME_API_*.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        taskId: { type: 'number', description: 'Leantime ticket id.' },
+        projectId: { type: 'number', description: 'Project id (for status resolution).' },
+        column: { type: 'string', enum: ['todo', 'doing', 'blocked', 'review', 'done'] },
+        status: { type: 'number', description: 'Explicit status id (bypasses column resolution).' },
+      },
+      required: ['taskId'],
+      additionalProperties: false,
+    },
+  },
+  {
+    name: 'byan_leantime_task_assign',
+    description: 'Set the assignee/editor of a Leantime task. Requires LEANTIME_API_*.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        taskId: { type: 'number', description: 'Leantime ticket id.' },
+        editorId: { type: 'number', description: 'Assignee/editor user id.' },
+      },
+      required: ['taskId', 'editorId'],
+      additionalProperties: false,
+    },
+  },
+  {
+    name: 'byan_leantime_task_get',
+    description: 'Fetch a single Leantime task by id. Requires LEANTIME_API_*.',
+    inputSchema: {
+      type: 'object',
+      properties: { taskId: { type: 'number', description: 'Leantime ticket id.' } },
+      required: ['taskId'],
+      additionalProperties: false,
+    },
+  },
+  {
+    name: 'byan_leantime_board_get',
+    description: "List a Leantime project's tasks grouped by lifecycle column. Requires LEANTIME_API_*.",
+    inputSchema: {
+      type: 'object',
+      properties: { projectId: { type: 'number', description: 'Leantime project id.' } },
+      required: ['projectId'],
+      additionalProperties: false,
+    },
+  },
 ];
 
 const server = new Server(
@@ -1309,33 +1405,6 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
       return { content: [{ type: 'text', text: JSON.stringify(result, null, 2) }] };
     }
 
-    if (name === 'byan_copilot_sessions') {
-      const result = listSessions({
-        limit: args.limit,
-        sinceIso: args.sinceIso,
-        cwdFilter: args.cwdFilter,
-      });
-      return { content: [{ type: 'text', text: JSON.stringify(result, null, 2) }] };
-    }
-
-    if (name === 'byan_copilot_session_events') {
-      const result = readSessionEvents({
-        sessionId: args.sessionId,
-        types: args.types,
-        limit: args.limit,
-      });
-      return { content: [{ type: 'text', text: JSON.stringify(result, null, 2) }] };
-    }
-
-    if (name === 'byan_copilot_search') {
-      const result = searchSessions({
-        query: args.query,
-        types: args.types,
-        limit: args.limit,
-      });
-      return { content: [{ type: 'text', text: JSON.stringify(result, null, 2) }] };
-    }
-
     if (name === 'byan_fd_start') {
       const state = fdStart({ featureName: args.featureName, force: args.force, strict: args.strict });
       return { content: [{ type: 'text', text: JSON.stringify(state, null, 2) }] };
@@ -1383,11 +1452,39 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
       };
     }
 
+    if (name === 'byan_insight_digest') {
+      const rootDir = process.env.CLAUDE_PROJECT_DIR || process.cwd();
+      const digest = harvestInsights({ rootDir });
+      return {
+        content: [
+          {
+            type: 'text',
+            text: JSON.stringify({ gated: true, digest, render: renderInsightDigest(digest) }, null, 2),
+          },
+        ],
+      };
+    }
+
+    if (name === 'byan_outcome_log') {
+      const line = validateForLog(args);
+      if (!line) {
+        return {
+          content: [{ type: 'text', text: JSON.stringify({ logged: false, reason: 'invalid_outcome' }) }],
+        };
+      }
+      const rootDir = process.env.CLAUDE_PROJECT_DIR || process.cwd();
+      const ok = appendOutcome(line, { rootDir });
+      return {
+        content: [{ type: 'text', text: JSON.stringify({ logged: ok, outcome: line }) }],
+      };
+    }
+
     if (name === 'byan_strict_lock_scope') {
       const r = strictLockScope({
         scopeText: args.scopeText,
         acceptanceCriteria: args.acceptanceCriteria,
         allowedPaths: args.allowedPaths,
+        domain: args.domain,
         force: args.force,
       });
       const st = strictGetStatus();
@@ -1432,6 +1529,18 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
     if (name === 'byan_strict_complete') {
       const r = strictComplete();
       const st = strictGetStatus();
+      // C3 learning loop: a completed strict session with an EXPLICIT ELO domain
+      // is a VALIDATED outcome. eloOutcomeForStrictComplete builds the line (the
+      // SAME helper the test exercises, so handler and test cannot drift); we
+      // append it to the buffer drain-advisory drains. The domain is the user's
+      // explicit lock_scope input, never inferred. Best-effort: a feed failure
+      // must not break completion.
+      try {
+        const eloLine = eloOutcomeForStrictComplete(r);
+        if (eloLine) appendOutcome(eloLine, { rootDir: process.env.CLAUDE_PROJECT_DIR || process.cwd() });
+      } catch {
+        // the learning feed must not break completion.
+      }
       const sync = await strictPushComplete({
         sessionId: st.strict_session_id,
         auditToken: r.audit_token,
@@ -1788,6 +1897,78 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
       return { content: [{ type: 'text', text: JSON.stringify(instructions, null, 2) }] };
     }
 
+    // ─── Leantime tools ───────────────────────────────────────────────
+    if (name === 'byan_leantime_ping') {
+      const status = {
+        api_url: process.env.LEANTIME_API_URL || null,
+        token_configured: Boolean(process.env.LEANTIME_API_TOKEN),
+        assign_user_configured: Boolean(process.env.LEANTIME_ASSIGN_USER_ID),
+        enabled: leantimeEnabled(),
+      };
+      if (status.enabled) {
+        const probe = await leantimeRpc(LEANTIME_METHODS.getAllProjects, {});
+        status.reachable = probe.ok;
+        if (!probe.ok) status.reason = probe.reason;
+        if (probe.hint) status.hint = probe.hint;
+      }
+      return { content: [{ type: 'text', text: JSON.stringify(status, null, 2) }] };
+    }
+
+    if (name === 'byan_leantime_project_ensure') {
+      requireLeantime();
+      const r = await leantimeEnsureProject({
+        name: args.name,
+        slug: args.slug,
+        clientId: args.clientId,
+        details: args.details,
+      });
+      return { content: [{ type: 'text', text: JSON.stringify(r, null, 2) }] };
+    }
+
+    if (name === 'byan_leantime_task_create') {
+      requireLeantime();
+      const r = await leantimeCreateTask({
+        projectId: args.projectId,
+        headline: args.headline,
+        description: args.description,
+        status: args.status,
+        priority: args.priority,
+        editorId: args.editorId,
+        tags: args.tags,
+        ...(args.type !== undefined ? { type: args.type } : {}),
+      });
+      return { content: [{ type: 'text', text: JSON.stringify(r, null, 2) }] };
+    }
+
+    if (name === 'byan_leantime_task_move') {
+      requireLeantime();
+      const r = await leantimeMoveTask({
+        taskId: args.taskId,
+        projectId: args.projectId,
+        column: args.column,
+        status: args.status,
+      });
+      return { content: [{ type: 'text', text: JSON.stringify(r, null, 2) }] };
+    }
+
+    if (name === 'byan_leantime_task_assign') {
+      requireLeantime();
+      const r = await leantimeAssignTask({ taskId: args.taskId, editorId: args.editorId });
+      return { content: [{ type: 'text', text: JSON.stringify(r, null, 2) }] };
+    }
+
+    if (name === 'byan_leantime_task_get') {
+      requireLeantime();
+      const r = await leantimeGetTask({ taskId: args.taskId });
+      return { content: [{ type: 'text', text: JSON.stringify(r, null, 2) }] };
+    }
+
+    if (name === 'byan_leantime_board_get') {
+      requireLeantime();
+      const r = await leantimeGetBoard({ projectId: args.projectId });
+      return { content: [{ type: 'text', text: JSON.stringify(r, null, 2) }] };
+    }
+
     throw new Error(`Unknown tool: ${name}`);
   } catch (err) {
     return {

package/install/templates/_byan/worker/launchers/README.md

@@ -19,7 +19,6 @@ Launcher workers are lightweight, single-purpose components that bridge the gap
 ```
 ┌─────────────────────────────────────────────────────────┐
 │                  USER INVOKES AGENT                     │
-│  gh copilot @bmad-agent-marc                            │
 │  claude --agent claude                                  │
 │  codex skill bmad-byan                                  │
 └─────────────┬───────────────────────────────────────────┘
@@ -54,18 +53,7 @@ Launcher workers are lightweight, single-purpose components that bridge the gap
 
 ## Workers
 
-### 1. launch-yanstaller-copilot.md
-
-**Platform:** GitHub Copilot CLI  
-**Icon:** 🤖  
-**Command:** `npx create-byan-agent`  
-**Called by:** `@bmad-agent-marc`
-
-**Purpose:** Launch yanstaller on Copilot CLI platform.
-
----
-
-### 2. launch-yanstaller-claude.md
+### 1. launch-yanstaller-claude.md
 
 **Platform:** Claude Code  
 **Icon:** 🎭  
@@ -79,7 +67,7 @@ Launcher workers are lightweight, single-purpose components that bridge the gap
 
 ---
 
-### 3. launch-yanstaller-codex.md
+### 2. launch-yanstaller-codex.md
 
 **Platform:** Codex/OpenCode  
 **Icon:** 📝  
@@ -107,7 +95,6 @@ Each worker has ONE task: Launch yanstaller command.
 ### Platform Hints
 Workers set environment variables to help yanstaller detect platform:
 ```bash
-BYAN_PLATFORM_HINT=copilot  # For Copilot CLI
 BYAN_PLATFORM_HINT=claude   # For Claude Code
 BYAN_PLATFORM_HINT=codex    # For Codex
 ```
@@ -166,7 +153,7 @@ Can be run multiple times safely.
 
 ## Separation of Concerns
 
-### Stub Agents (marc/claude/codex)
+### Stub Agents (claude/codex)
 - Detect invocation
 - Call launcher worker
 - Minimal logic
@@ -185,7 +172,6 @@ Can be run multiple times safely.
 - Platform-specific integration
 - MCP server creation (Claude)
 - Skill file creation (Codex)
-- GitHub agent installation (Copilot)
 
 ---
 
@@ -196,7 +182,6 @@ _byan/
 └── workers/
     └── launchers/
         ├── README.md (this file)
-        ├── launch-yanstaller-copilot.md
         ├── launch-yanstaller-claude.md
         └── launch-yanstaller-codex.md
 ```
@@ -207,9 +192,6 @@ _byan/
 
 ### Manual Test
 ```bash
-# Test Copilot launcher
-node -e "require('./_byan/worker/launchers/worker-launch-yanstaller-copilot').launch()"
-
 # Test Claude launcher
 node -e "require('./_byan/worker/launchers/worker-launch-yanstaller-claude').launch()"
 
@@ -219,7 +201,7 @@ node -e "require('./_byan/worker/launchers/worker-launch-yanstaller-codex').laun
 
 ### Expected Output
 ```
-🤖 Launching Yanstaller on Copilot CLI...
+🎭 Launching Yanstaller on Claude Code...
 [Yanstaller interview UI appears]
 ```
 
@@ -287,7 +269,6 @@ install/
         └── workers/
             └── launchers/
                 ├── README.md
-                ├── launch-yanstaller-copilot.md
                 ├── launch-yanstaller-claude.md
                 └── launch-yanstaller-codex.md
 ```
@@ -297,7 +278,6 @@ install/
 ## Version History
 
 - **1.0.0** (2026-02-10): Initial release
-  - Copilot launcher
   - Claude launcher
   - Codex launcher
 

package/install/templates/_byan/worker/workers.md

@@ -125,7 +125,6 @@ if (complexityScore < 30) {
 **Utilisation :** Lancer yanstaller sur chaque plateforme
 
 **Fichiers :**
-- `_byan/worker/launchers/launch-yanstaller-copilot.md`
 - `_byan/worker/launchers/launch-yanstaller-claude.md`
 - `_byan/worker/launchers/launch-yanstaller-codex.md`
 
@@ -287,24 +286,25 @@ very different optimal targets depending on whether they run **alongside
 siblings** (parallel) or **in sequence**. The v2 router adds a
 `parallelizable` axis and emits an **execution strategy**, not a model.
 
-Implementation : `src/core/dispatcher/execution-router.js` and the MCP
-tool `byan_dispatch` (both share the same table).
+Implementation : the MCP tool `byan_dispatch`
+(`_byan/mcp/byan-mcp-server/lib/dispatch.js`), the single source of truth. The
+strategy comes from the score + `parallelizable` ; the model tier is a separate
+axis, derived from the task NATURE via `native-tiers.js`.
 
 ```
 score < 15                           → main-thread
 score 15-39 + parallelizable: true   → agent-subagent-worktree
-score 15-39 + parallelizable: false  → mcp-worker-haiku
-score >= 40                          → main-thread-opus
+score 15-39 + parallelizable: false  → mcp-worker
+score >= 40                          → main-thread (heavy)
 ```
 
 Rationale :
 
 | Strategy | When | Why |
 |---|---|---|
-| `main-thread` | Trivial task | Spawning anything costs more than solving inline. |
+| `main-thread` | Trivial or heavy task | Spawning costs more than solving inline (trivial), or the work is heavy and stays in the main thread. |
 | `agent-subagent-worktree` | Medium parallel | Claude Code Agent tool with `isolation: "worktree"` amortizes boot cost across the wall-clock savings. |
-| `mcp-worker-haiku` | Medium sequential | Delegate to a lightweight Haiku via MCP tool — no subagent boot, cheaper than main thread. |
-| `main-thread-opus` | Complex | Reasoning depth needed; subagent boot + context handoff would waste more than the delegation saves. |
+| `mcp-worker` | Medium sequential | Delegate to a worker via MCP tool — no subagent boot, cheaper than the main thread. The model tier is set separately, by nature. |
 
 The score threshold of 15 is where Claude Code `Agent` tool boot overhead
 (~5-10k tokens for system prompt + tools) stops being worth it for
@@ -478,7 +478,6 @@ _byan/
 └── workers/
     └── launchers/
         ├── README.md
-        ├── launch-yanstaller-copilot.md
         ├── launch-yanstaller-claude.md
         └── launch-yanstaller-codex.md
 

package/install/templates/_byan/workflow/simple/bmb/byan-benchmark/workflow.md

@@ -0,0 +1,86 @@
+---
+name: byan-benchmark
+description: 'DATA-only benchmark engine for any decision fork: options x weighted-criteria matrix + best-first reco + dissent. Markdown fallback for non-native platforms (dual-path).'
+---
+
+# byan-benchmark Workflow (markdown fallback)
+
+**Goal:** Given a decision fork (>=2 non-substitutable options + weighted
+criteria + an optional judge panel), produce a scored options-x-criteria matrix,
+a best-first recommendation, and the dissenting view - as DATA. The human gate
+and the rendered table live in the orchestrating `byan-benchmark` skill, not here.
+
+**Your Role:** You are the benchmark engine. You score; the user decides. State
+mutations (FD/strict) stay out of this workflow - that is the skill's job at the
+gate.
+
+This markdown is the dual-path FALLBACK. The native engine is
+`.claude/workflows/byan-benchmark.js`; `resolveWorkflow('byan-benchmark')`
+prefers the `.js` and falls back to this file on platforms without the native
+Workflow tool.
+
+---
+
+## ARGS CONTRACT
+
+- `question` - the fork stated as a question.
+- `options` - array of `{ name, note? }` (>=2 for a real benchmark).
+- `criteria` - array of `{ name, weight }` (>=1).
+- `judges` - optional reusable panel `[{ key, lens, weighting }]`; default a
+  single neutral judge.
+- `domain` - drives strict floors (`security`/`performance` -> L2, `compliance`
+  -> L1).
+- `scope` - `internal` (no external links, coherence-first) or `external`
+  (sourcing allowed, but a URL only if opened this turn).
+
+---
+
+## STEPS
+
+### 1. RECON - parse the fork
+
+Normalise `options` to `[{name, note?}]` and `criteria` to `[{name, weight}]`
+(default weight 1). The fork is **valid** only if there are >=2 distinct,
+non-substitutable options AND >=1 criterion. A degenerate / obvious-default fork
+is not benchmarkable - return `degenerate: true` with a reason so the skill emits
+a `BYAN-BENCH:skip` marker.
+
+### 2. SOURCE - gather evidence per option
+
+For each option, write one evidence note per criterion. Routing decides links
+before depth: `internal` stays on model-knowledge with no external links;
+`external` may cite a source, but a URL appears only if WebFetch opened it this
+turn - otherwise the claim is `unverified: true`. Honour the strict domain floor.
+
+### 3. JUDGE - score each cell
+
+Per the judge panel (default neutral), score each option on each criterion 1-10,
+grade the evidence level against the 5-level rubric (L1 95% spec -> L5 20%
+opinion), and compute `weightedTotal = sum(score * weight)`. A cell below the
+strict-domain floor is flagged `unverified: true`.
+
+### 4. RECOMMEND - rank best-first + dissent
+
+Consolidate the judges into one matrix, best-first by combined weighted total.
+Recommend the winner with a one-line best-first reco. Use `confidence: assertive`
+only when the winner leads by a wide margin and its key cells are verified;
+otherwise use `confidence: lean` (low-confidence, hedged verb). Capture the
+dissent: the runner-up a reasonable judge would defend and the criterion it wins on.
+
+---
+
+## RETURN (DATA only)
+
+```json
+{ "workflow": "byan-benchmark", "question": "...", "scope": "internal",
+  "domain": "general", "options": [...], "criteria": [...],
+  "matrix": [{ "option": "...", "cells": [{ "criterion": "...", "verdict": "...",
+               "level": "L2", "score": 8, "source": "...", "unverified": false }],
+             "total": 0 }],
+  "recommendation": { "best": "...", "line": "...", "confidence": "assertive|lean" },
+  "dissent": { "option": "...", "why": "..." },
+  "degenerate": false, "needsHumanGate": true }
+```
+
+No state mutation. No emoji. The skill renders the compact 1-table, emits the
+BYAN-BENCH marker, and records state via MCP at the human gate.

package/install/templates/_byan/workflow/simple/byan/feature-workflow.md

@@ -111,8 +111,8 @@ INIT
 |------------------|-------|-----------|
 | < 15 | `main-thread` | Inline dans le contexte courant, zéro overhead de délégation |
 | < 40 + parallélisable | `agent-subagent-worktree` | Agent tool Claude Code avec isolation worktree |
-| < 40 séquentiel | `mcp-worker-haiku` | Worker Haiku léger via MCP |
-| ≥ 40 | `main-thread-opus` | Garde en main thread, raisonnement Opus |
+| < 40 séquentiel | `mcp-worker` | Worker léger via MCP (le tier de modèle vient de la nature, pas de la taille) |
+| ≥ 40 | `main-thread` | Garde en main thread (lourd) ; modèle hérité de la session |
 
 > Le score (0-100) est estimé depuis la complexité de la tâche (longueur si absent). Appeler `byan_dispatch` pour le calcul — ne pas réinventer les seuils ici.