@dotdrelle/wiki-manager 0.15.66 → 0.15.71

package/.env.example

@@ -39,11 +39,8 @@ WORKSPACES_ROOT=/path/to/workspaces
 #   "documents": { "headers": { "Authorization": "Bearer <DOCUMENTS_MCP_AUTH_TOKEN>" } }
 #
 # Leave empty to disable authentication on that agent (not recommended in production).
-
 CME_MCP_AUTH_TOKEN=
 DOCUMENTS_MCP_AUTH_TOKEN=
-# Generated by `wiki-workspace agents up` when missing.
-CONNECTORS_MCP_AUTH_TOKEN=
 
 # ── Agent ports (optional, change only if defaults conflict) ───────────────────
 
@@ -55,6 +52,16 @@ CONNECTORS_MCP_AUTH_TOKEN=
 #
 # Enable the opt-in agent-connectors service:
 CONNECTORS_ENABLED=false
+CONNECTORS_MCP_AUTH_TOKEN=
+
+# Agentic runtime gateway (Deep Agents) — enabled by default, started by
+# `agents up`. The manager never starts it, only discovers it via
+# agent-runtimes.json; set GATEWAY_ENABLED=false to opt out.
+GATEWAY_ENABLED=true
+# GATEWAY_PORT=7789
+# Generated by `agents up` if empty, like the other agent tokens; the manager
+# sends it as the Bearer header on every gateway call.
+GATEWAY_AUTH_TOKEN=
 #
 # The wikiLLM Google OAuth application is baked into the agent-connectors image
 # at build time, from agent-external/agent-connectors/.env.build.local. Neither

package/README.md

@@ -120,6 +120,62 @@ in isolated workspaces.
 
 ![wikiLLM functional diagram — inputs, MCP calls and outputs around the agentic orchestrator and workspaces](https://raw.githubusercontent.com/dotdrelle/llm-wiki-manager/main/docs/architecture.png)
 
+## How wikiLLM compares
+
+Several open projects now build a Markdown wiki with an LLM. They target
+**different problems** — the useful questions are *what goes in, what comes out,
+and who operates it*. Snapshot as of 2026; all of these move quickly.
+
+✅ first-class · 🟡 partial or indirect · ❌ not a goal
+
+Projects compared: [OpenWiki](https://github.com/langchain-ai/openwiki),
+[DeepWiki-Open](https://github.com/asyncfuncai/deepwiki-open),
+[GraphRAG](https://github.com/microsoft/graphrag).
+
+<table>
+<thead>
+<tr><th><small>Need</small></th><th><small><strong>wikiLLM</strong></small></th><th><small><strong>OpenWiki</strong></small></th><th><small><strong>DeepWiki-Open</strong></small></th><th><small><strong>GraphRAG</strong></small></th></tr>
+</thead>
+<tbody>
+<tr><td><small>Input</small></td><td><small>✅ Business docs — Confluence, PDF, Office, SaaS</small></td><td><small>✅ Codebase (code mode)</small></td><td><small>✅ Code repo → diagrams</small></td><td><small>🟡 Plain-text corpus only</small></td></tr>
+<tr><td><small>Output</small></td><td><small>✅ Maintained wiki <strong>+ deliverables</strong> from your templates</small></td><td><small>🟡 Wiki about the code, for agents</small></td><td><small>🟡 Interactive wiki + diagrams</small></td><td><small>❌ Entity graph + summaries — no wiki</small></td></tr>
+<tr><td><small>Keep current</small></td><td><small>✅ Re-ingest / scheduled</small></td><td><small>✅ <code>--update</code>, CI action</small></td><td><small>🟡 Regenerated per run</small></td><td><small>✅ <code>graphrag update</code> (delta)</small></td></tr>
+<tr><td><small>Evidence &amp; citations</small></td><td><small>🟡 Cites retrieved context</small></td><td><small>✅ Claims tied to versioned source</small></td><td><small>🟡 RAG-cited answers</small></td><td><small>✅ Citations to text units</small></td></tr>
+<tr><td><small>Corpus-wide Q&amp;A</small></td><td><small>🟡 BM25 + vector feeding generation</small></td><td><small>❌</small></td><td><small>🟡 Repo-scoped RAG chat</small></td><td><small>✅ Local/global community search</small></td></tr>
+<tr><td><small>Team UI</small></td><td><small>✅ Web console — wiki, graph, chat, runs (single-user today)</small></td><td><small>🟡 Local viewer + CLI chat</small></td><td><small>✅ Self-hosted web app + RAG</small></td><td><small>❌ Library / CLI</small></td></tr>
+<tr><td><small>Multi-project isolation</small></td><td><small>✅ Workspaces, own services, ports, secrets</small></td><td><small>❌ One wiki per run</small></td><td><small>❌ One wiki per repo</small></td><td><small>❌ One index per corpus</small></td></tr>
+<tr><td><small>Orchestration &amp; governance</small></td><td><small>✅ Approval-gated dispatcher, budgets, idempotent writes, crash recovery</small></td><td><small>❌ One agent loop</small></td><td><small>❌ One generation pipeline</small></td><td><small>❌ Indexing pipeline</small></td></tr>
+<tr><td><small>Connectors as services</small></td><td><small>✅ Independent MCP agents (Confluence, docs, e-mail…)</small></td><td><small>🟡 Built-in connector set</small></td><td><small>❌</small></td><td><small>❌</small></td></tr>
+<tr><td><small>Offline / local models</small></td><td><small>✅ Per-workspace OpenAI-compatible or gateway (Ollama, vLLM, MLX…)</small></td><td><small>✅ 13+ providers</small></td><td><small>✅ Ollama</small></td><td><small>✅ Any OpenAI-compatible</small></td></tr>
+<tr><td><small>License</small></td><td><small>❌ PolyForm <strong>Noncommercial</strong></small></td><td><small>✅ MIT</small></td><td><small>✅ MIT</small></td><td><small>✅ MIT</small></td></tr>
+</tbody>
+</table>
+
+**The short version:**
+
+- **OpenWiki** and **DeepWiki-Open** document *source code*. Point wikiLLM at a
+  repository and there is nothing for it to ingest; point either of them at a
+  stack of Confluence pages and a Word document and that is not their job.
+- **GraphRAG** builds *retrieval structure*, not a wiki you read or deliverables
+  you ship — it is a strong back end for corpus-wide Q&A, and complementary
+  rather than competing.
+- **wikiLLM** is the only one of the four whose output is *both* a browsable wiki
+  *and* regenerated business documents, and the only one with the operational
+  layer — isolated projects, bounded approvals, automatic recovery, a web
+  console — that a shared internal tool needs.
+
+**What wikiLLM does *not* try to do (today):**
+
+- Document a codebase for coding agents — that is OpenWiki / DeepWiki-Open
+  territory.
+- Serve a true multi-user instance with per-user identity and an attributed
+  audit trail. This is a single-user deployment baseline (see the scope note
+  above); multi-user is specified and planned next.
+- Expose a graph-query API over the corpus the way GraphRAG does; retrieval is
+  BM25 plus a vector index feeding generation.
+- Ship or host the multi-provider AI gateway — routing to several providers is
+  supported, the gateway itself is infrastructure you bring.
+
 ## Quick start — your first wiki in ~5 minutes
 
 The fastest way in: a **browsable wiki, its dependency graph, and a grounded
@@ -402,6 +458,7 @@ answer "what is this and how do I start it", and stop there.
 | [`docs/configuration.md`](docs/configuration.md) | Every configuration key: root `.env`, Compose overrides, `mcp.endpoints.json`, workspace `.env`, `.wikirc.yaml`, parallelism |
 | [`docs/technical-reference.md`](docs/technical-reference.md) | Workspace model, services, the `donna` shell, agent tooling, orchestration and activity contracts, security model |
 | [`docs/authoring-skills.md`](docs/authoring-skills.md) | Writing a workspace skill: what splits a body into runs, chains, concurrency, parameters, and the interpretation rules |
+| [`docs/agentic-runtime.md`](docs/agentic-runtime.md) | The external agentic runtime: `agent-runtimes.json`, the `RuntimeProvider` contract, the gateway, and governance |
 | [`docs/claude-desktop.md`](docs/claude-desktop.md) | Using a workspace from Claude Desktop |
 | [`CLAUDE.md`](CLAUDE.md) | Repository guidance: invariants to preserve when changing this code |
 

package/agent-runtimes.example.json

@@ -0,0 +1,68 @@
+{
+  "_comment": "Agentic runtimes for this manager (RFC § 37), seeded to agent-runtimes.json on first scaffold — enabled by default, matching the scaffold's GATEWAY_ENABLED=true (set GATEWAY_ENABLED=false in .env, or enabled:false here, to opt out). Rules: (1) a capability is read-only unless it declares mutationClass or defaultRequiresApproval; (2) the 'plan' operation is always a dry-run and never pauses for approval; (3) read/write pairs are TWO capabilities, not two operations, so the approval class stays per-capability; (4) the runtime has eyes (read tools), ideas (free reasoning) and a mouth (gated side-effects), but no hands on the workspace — the hands are the DAG. See llm-wiki-manager/docs/agentic-runtime.md.",
+  "runtimes": [
+    {
+      "id": "deepagents",
+      "type": "deepagents",
+      "endpoint": "http://localhost:7789",
+      "enabled": true,
+      "capabilities": [
+        {
+          "name": "agent.review",
+          "operations": ["run"],
+          "description": "Read-only audit of a wiki workspace: compare source documents against the existing concept pages, identify missing or under-covered classes, and produce a structured gap report. No mutation.",
+          "aliases": ["audit", "review", "analyze", "compare", "check"]
+        },
+        {
+          "name": "agent.consistency",
+          "operations": ["run"],
+          "description": "Read-only detection of contradictions and inconsistencies between wiki pages and their sources, citing the conflicting passages. No mutation.",
+          "aliases": ["consistency", "coherence", "contradictions", "conflicts"]
+        },
+        {
+          "name": "agent.synthesize",
+          "operations": ["run"],
+          "description": "Read-only cross-source synthesis: build a structured summary of several documents on one subject, using web search tools when available to complement the sources. No mutation.",
+          "aliases": ["synthesize", "summarize", "synthesis"]
+        },
+        {
+          "_comment": "operations: 'plan' reports the proposal only (no pause, dry-run); 'run' integrates it and waits for human approval (defaultRequiresApproval).",
+          "name": "agent.plan",
+          "operations": ["plan", "run"],
+          "description": "Given an analysis, propose a deterministic DAG plan (ingest, build, sync) as a validated fragment. The 'plan' operation only reports the proposal; 'run' integrates it and waits for human approval.",
+          "aliases": ["plan", "propose"],
+          "aliasOperations": { "plan": "plan", "propose": "plan", "apply": "run", "execute": "run" },
+          "defaultRequiresApproval": true
+        },
+        {
+          "name": "agent.answer",
+          "operations": ["run"],
+          "description": "Read-only research answer: investigate a question using the wiki sources and web search, and reply with a grounded answer. No mutation.",
+          "aliases": ["answer", "question", "explain"]
+        },
+        {
+          "_comment": "mutationClass 'ingest': writes the findings into the workspace inbox, approval required.",
+          "name": "agent.research",
+          "operations": ["run"],
+          "description": "Web research grounded in the wiki: answer a complex question using wiki sources and web search, then write the findings into the workspace inbox. Mutation, approval required.",
+          "aliases": ["research", "investigate"],
+          "mutationClass": "ingest"
+        },
+        {
+          "name": "agent.preview",
+          "operations": ["run"],
+          "description": "Compose the notification report from the workspace profile and show it, without sending. No mutation.",
+          "aliases": ["preview", "draft", "compose"]
+        },
+        {
+          "_comment": "defaultRequiresApproval: sending the mail pauses for human approval.",
+          "name": "agent.notify",
+          "operations": ["run"],
+          "description": "Read the workspace profile for the notification recipient, then send the report by email. Mutation, approval required.",
+          "aliases": ["notify", "send", "email", "report"],
+          "defaultRequiresApproval": true
+        }
+      ]
+    }
+  ]
+}

package/agents.docker-compose.yml

@@ -25,6 +25,12 @@
 #   DOCUMENT_LLM_MODEL     — lightonai/LightOnOCR-2-1B
 #   DOCUMENT_LLM_API_KEY   — OpenAI/OpenAI-compatible API key for document OCR
 #   DOCUMENT_LLM_TIMEOUT_SECONDS — 120
+#   GATEWAY_PORT           — 7789
+#   GATEWAY_ENABLED        — false; true to start the agentic runtime gateway
+#   GATEWAY_AUTH_TOKEN     — bearer token for the gateway (generated if empty);
+#                            the manager sends it on every call. The model is
+#                            NOT configured here: the manager sends its active
+#                            profile model with every run.
 #
 # Set these variables in a .env file at the directory where you run wiki-workspace,
 # or export them in your shell before running wiki-workspace agents up.
@@ -62,7 +68,7 @@ services:
     restart: unless-stopped
 
   documents:
-    image: dotdrelle/agent-wiki-documents:latest
+    image: dotdrelle/agent-documents:latest
     user: "${UID:-1000}:${GID:-1000}"
     ports:
       - "${DOCUMENTS_MCP_PORT:-3337}:8080"
@@ -134,3 +140,35 @@ services:
       - ${AGENTS_DATA_DIR:-./.agents-data}/connectors:/data
       - ${WORKSPACES_ROOT:?Set WORKSPACES_ROOT to the directory containing all workspace folders}:/workspaces
     restart: unless-stopped
+
+  # Agentic runtime gateway — the external Deep Agents service the manager
+  # routes open-ended analysis to (agent-runtimes.json, endpoint 7789).
+  # Opt-in like connectors: set GATEWAY_ENABLED=true. The manager never starts
+  # this service, only discovers it; its failure leaves every DAG untouched.
+  gateway:
+    profiles: [gateway]
+    image: dotdrelle/wiki-agentic-gateway:latest
+    user: "${UID:-1000}:${GID:-1000}"
+    ports:
+      - "${GATEWAY_PORT:-7789}:7789"
+    environment:
+      - GATEWAY_CONFIG_DIR=/config
+      - GATEWAY_AUTH_TOKEN=${GATEWAY_AUTH_TOKEN:-}
+      - NODE_USE_ENV_PROXY=${NODE_USE_ENV_PROXY:-}
+      - HTTPS_PROXY=${HTTPS_PROXY:-}
+      - HTTP_PROXY=${HTTP_PROXY:-}
+      - NO_PROXY=${NO_PROXY:-localhost,127.0.0.1,host.docker.internal}
+    extra_hosts:
+      - host.docker.internal:host-gateway
+    volumes:
+      # One file, two readers: the manager routes on agent-runtimes.json, the
+      # gateway serves the capabilities of its own entry from the same file.
+      # Absolute path, computed by wiki-workspace from the manager state dir.
+      # Declared AFTER the /config data mount: for overlapping bind mounts the
+      # child mount wins, and the reverse order silently hid the file under
+      # the directory — the gateway degraded to its built-in default
+      # capabilities on every start.
+      - ${AGENTS_DATA_DIR:-./.agents-data}/gateway:/config
+      - ${AGENT_RUNTIMES_FILE:?Set AGENT_RUNTIMES_FILE}:/config/agent-runtimes.json:ro
+      - ${WORKSPACES_ROOT:?Set WORKSPACES_ROOT to the directory containing all workspace folders}:/workspaces
+    restart: unless-stopped

package/docker-compose.yml

@@ -110,10 +110,10 @@ services:
       - host.docker.internal:host-gateway
     restart: unless-stopped
 
-  # ── agent-wiki-production ─────────────────────────────────────────────────
+  # ── agent-production ─────────────────────────────────────────────────
 
   production-mcp:
-    image: dotdrelle/agent-wiki-production:latest
+    image: dotdrelle/agent-production:latest
     user: "${UID:-1000}:${GID:-1000}"
     labels:
       wiki-manager.description: "Production MCP server for ingest/build/export jobs."
@@ -130,7 +130,7 @@ services:
       # error. Every compose-deployed ingest then ran without the Lot 4 barrier
       # and left the published map stale — the very defect that work fixed.
       # `copy` stays out on purpose: it is the legacy step, opt-in only.
-      - PRODUCTION_ALLOWED_STEPS=${PRODUCTION_ALLOWED_STEPS:-doctor,ingest,ingest_plan,ingest_apply,concepts,reclassify-concepts,taxonomy,build,export,polish,restore,pipeline}
+      - PRODUCTION_ALLOWED_STEPS=${PRODUCTION_ALLOWED_STEPS:-doctor,ingest,ingest_plan,ingest_apply,build,export,polish,restore,pipeline}
       - PRODUCTION_REQUIRE_CONFIRMATION=${PRODUCTION_REQUIRE_CONFIRMATION:-false}
       # Parallelism levers — effective concurrency ≈ recommendedConcurrency.
       # Intermediate defaults (4/8). Low profile 2/4, high profile 8/16.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dotdrelle/wiki-manager",
-  "version": "0.15.66",
+  "version": "0.15.71",
   "description": "Agentic shell and orchestration cockpit for llm-wiki workspaces.",
   "repository": {
     "type": "git",
@@ -19,7 +19,7 @@
   },
   "scripts": {
     "start": "bun ./bin/wiki-manager.js",
-    "test": "node --test src/core/skillInvocation.test.js src/core/skillCompiler.test.js src/runtime/skillRun.test.js src/runtime/controlDrain.test.js src/runtime/controlCancellation.test.js src/cli/runtimeStartup.test.js src/cli/wiki-manager.test.js src/agent/graph.test.js src/agent/skillRecursion.test.js src/contracts/schemas.test.js src/core/activity.test.js src/core/env.test.js src/core/agentsCompose.test.js src/core/profileServiceStatus.test.js src/core/workspaceProfile.test.js src/core/buildInfo.test.js src/core/agentEvents.test.js src/core/skillChainView.test.js src/core/runtimeLog.test.js src/activity/activityAggregator.test.js src/graph/runGraphProjector.test.js src/core/workflow.test.js src/core/planPatch.test.js src/core/agentLoop.test.js src/core/plan.test.js src/core/mcp.test.js src/core/toolLoop.test.js src/core/documentIntake.test.js src/core/dockerCompose.test.js src/core/otherWorkspacesRunning.test.js src/core/wikiSetup.test.js src/core/wikiWorkspace.test.js src/core/wikirc.test.js src/core/workspaceInherit.test.js src/core/cacert.test.js src/core/composeOverrides.test.js src/core/setEnvValue.test.js src/core/commandFailure.test.js src/core/currentArtifact.test.js src/core/googleGrants.test.js src/core/modelFetch.test.js src/core/startupCheck.test.js src/core/queueStore.test.js src/orchestrator/agentRegistry.test.js src/orchestrator/capabilityRegistry.test.js src/orchestrator/capabilityResolver.test.js src/orchestrator/planValidator.test.js src/orchestrator/planIntegrator.test.js src/orchestrator/taskStatuses.test.js src/orchestrator/scheduler.test.js src/orchestrator/attemptManager.test.js src/orchestrator/resultAggregator.test.js src/orchestrator/approvalPolicy.test.js src/orchestrator/dispatcher.test.js src/orchestrator/objectiveResolver.test.js src/commands/slash.test.js src/shell/repl.test.js src/shell/setupWizardModality.test.js src/shell/setupWizardPlaceholders.test.js src/shell/setupWizardSuggestions.test.js src/shell/setupWizardDiscovery.test.js src/shell/wrapText.test.js src/runtime/lifecycle.test.js src/runtime/store.test.js src/runtime/workspaceIsolation.test.js src/runtime/controlMessages.test.js src/runtime/recoveryManager.test.js src/runtime/server.test.js src/runtime/supervisor.test.js src/runtime/delegation.test.js src/runtime/runner.test.js src/runtime/runner.e2e.test.js src/runtime/skillChain.e2e.test.js src/runtime/donna-contract.test.js src/runtime/approvals.test.js src/runtime/auth.test.js",
+    "test": "node --test src/core/skillInvocation.test.js src/core/skillCompiler.test.js src/runtime/skillRun.test.js src/runtime/controlDrain.test.js src/runtime/controlCancellation.test.js src/cli/runtimeStartup.test.js src/cli/wiki-manager.test.js src/agent/graph.test.js src/agent/skillRecursion.test.js src/contracts/schemas.test.js src/core/activity.test.js src/core/env.test.js src/core/agentsCompose.test.js src/core/profileServiceStatus.test.js src/core/workspaceProfile.test.js src/core/buildInfo.test.js src/core/agentEvents.test.js src/core/skillChainView.test.js src/core/runtimeLog.test.js src/core/runtimeEventAdapter.test.js src/activity/activityAggregator.test.js src/graph/runGraphProjector.test.js src/core/workflow.test.js src/core/planPatch.test.js src/core/agentLoop.test.js src/core/plan.test.js src/core/mcp.test.js src/core/toolLoop.test.js src/core/documentIntake.test.js src/core/dockerCompose.test.js src/core/otherWorkspacesRunning.test.js src/core/wikiSetup.test.js src/core/wikiWorkspace.test.js src/core/wikirc.test.js src/core/workspaceInherit.test.js src/core/cacert.test.js src/core/composeOverrides.test.js src/core/setEnvValue.test.js src/core/commandFailure.test.js src/core/currentArtifact.test.js src/core/googleGrants.test.js src/core/modelFetch.test.js src/core/startupCheck.test.js src/core/queueStore.test.js src/orchestrator/agentRegistry.test.js src/orchestrator/capabilityRegistry.test.js src/orchestrator/capabilityResolver.test.js src/orchestrator/planValidator.test.js src/orchestrator/planIntegrator.test.js src/orchestrator/taskStatuses.test.js src/orchestrator/scheduler.test.js src/orchestrator/attemptManager.test.js src/orchestrator/resultAggregator.test.js src/orchestrator/approvalPolicy.test.js src/orchestrator/dispatcher.test.js src/orchestrator/objectiveResolver.test.js src/orchestrator/providers/fakeRuntimeProvider.test.js src/orchestrator/providers/runtimeProviders.test.js src/orchestrator/providers/dispatcherExternalRuntime.test.js src/orchestrator/providers/deepAgentsProvider.test.js src/commands/slash.test.js src/shell/repl.test.js src/shell/setupWizardModality.test.js src/shell/setupWizardPlaceholders.test.js src/shell/setupWizardSuggestions.test.js src/shell/setupWizardDiscovery.test.js src/shell/wrapText.test.js src/runtime/lifecycle.test.js src/runtime/store.test.js src/runtime/workspaceIsolation.test.js src/runtime/controlMessages.test.js src/runtime/recoveryManager.test.js src/runtime/server.test.js src/runtime/supervisor.test.js src/runtime/delegation.test.js src/runtime/runner.test.js src/runtime/runner.e2e.test.js src/runtime/skillChain.e2e.test.js src/runtime/donna-contract.test.js src/runtime/approvals.test.js src/runtime/auth.test.js",
     "check-versions": "node scripts/check-versions.js",
     "prepack": "node scripts/check-versions.js",
     "prepublishOnly": "node scripts/check-versions.js",
@@ -38,6 +38,7 @@
     "docker-compose.override.example.yml",
     "agents.docker-compose.override.example.yml",
     "mcp.endpoints.example.json",
+    "agent-runtimes.example.json",
     ".env.example",
     "tsconfig.json",
     "bunfig.toml",

package/src/activity/activityAggregator.test.js

@@ -144,12 +144,12 @@ test('aggregateActivity keeps a healthy active task out of the error color when
       key: 'activity-b',
       status: 'running',
       terminal: false,
-      progress: { percent: 35, stepId: 'running-b', label: 'Ingest application-orea.md', detail: 'LLM running' },
+      progress: { percent: 35, stepId: 'running-b', label: 'Ingest sample-application.md', detail: 'LLM running' },
     }],
   }, []);
 
   const line = aggregated.lines[0];
   assert.equal(line.status, '35 %');
   assert.match(line.label, /^\[\.\.\.\]/);
-  assert.equal(line.progress.label, 'Ingest application-orea.md');
+  assert.equal(line.progress.label, 'Ingest sample-application.md');
 });

package/src/agent/graph.js

@@ -891,7 +891,7 @@ export async function handleRuntimeControlTool(session, tool, args = {}) {
           delegated: true,
           runId: inRun.runId,
           summary: inRun.summary ?? null,
-          message: `Action lancée (${String(inRun.runId).slice(0, 8)}) après validation du plan réel : ${inRun.summary?.tasks ?? 0} tâche(s), ${inRun.summary?.agent ?? 'agent résolu'}. Exécution en cours.`,
+          message: `Action started (${String(inRun.runId).slice(0, 8)}) after real-plan validation: ${inRun.summary?.tasks ?? 0} task(s), ${inRun.summary?.agent ?? 'resolved agent'}. Execution in progress.`,
         });
       }
       const result = await postRuntimeDelegate(objective, { url, workspace });
@@ -900,9 +900,9 @@ export async function handleRuntimeControlTool(session, tool, args = {}) {
             delegated: true,
             runId: result.runId,
             summary: result.delegation ?? null,
-            message: `Action lancée (${String(result.runId).slice(0, 8)}) après validation du plan réel : ${result.delegation?.tasks ?? 0} tâche(s), ${result.delegation?.agent ?? 'agent résolu'}. Exécution en cours.`,
+            message: `Action started (${String(result.runId).slice(0, 8)}) after real-plan validation: ${result.delegation?.tasks ?? 0} task(s), ${result.delegation?.agent ?? 'resolved agent'}. Execution in progress.`,
           })
-        : `Délégation refusée : ${result?.error ?? JSON.stringify(result)}`;
+        : `Delegation refused: ${result?.error ?? JSON.stringify(result)}`;
     }
     if (tool === 'run_skill') {
       const skillName = String(args.skillName ?? '').trim();
@@ -952,12 +952,13 @@ export async function handleRuntimeControlTool(session, tool, args = {}) {
        par ressemblance.
 
        La garde de cycle ci-dessus ne voit que les répétitions. Elle laissait
-       donc passer la cascade réellement observée sur un `/wiki-ingest` :
-       l'intention n°2 du corps est mot pour mot celui de
-       `/wiki-rebuild-concepts`, dont l'intention décrit à son tour
-       `/wiki-reclassify`, puis `/wiki-taxonomy`. Trois compétences distinctes,
-       aucun cycle, et la grille de concepts comme la taxonomie reconstruites
-       plusieurs fois pour une seule demande.
+       donc passer la cascade réellement observée sur un `/wiki-ingest` : une
+       intention compilée décrivait mot pour mot le corps d'une compétence
+       sœur, dont l'intention décrivait à son tour la suivante. Trois
+       compétences distinctes, aucun cycle, et la grille de concepts comme la
+       taxonomie reconstruites plusieurs fois pour une seule demande.
+       (Ces compétences sœurs ont disparu avec la simplification 0.15.66 ; la
+       garde, elle, reste.)
 
        Une intention compilée EST déjà le travail à faire : elle se délègue.
        La composition explicite reste ouverte — un corps qui nomme sa cible dit
@@ -1247,6 +1248,7 @@ export function buildAgentSystemPrompt(state) {
     'When an action fails or is refused for lack of an authorization grant or scope (rather than a missing capability), say exactly that and name the primitive that grants it. Do not describe the feature as unavailable.',
     'For an action with no matching direct tool, call runtime__delegate with the user objective only. The runtime chooses the capability, operation, agent and plan, including a validated single task for executor-only agents. Never choose those identifiers yourself. Never call <provider>__agent_plan, <provider>__agent_execute, legacy production__production_start_job, wiki__plan_set, or wiki__plan_done from interactive chat.',
     'Do not ask the user which sources, files, connectors, or templates to use for an ingest, build, or export: the specialized agent discovers them from the workspace. When the objective is clear (e.g. "lance une ingestion"), delegate it as stated, without a clarifying question.',
+    'The concept map is not a deliverable and has no build, rebuild, reclassify, group or taxonomy pass of its own. Concepts are the folders produced by ingestion itself (wiki/concepts/<concept>/<subject>.md — the concept IS the folder), and the /graph taxonomy derives from them. So "rebuild / refresh / redo the concepts" means run an ingestion (the wiki-ingest skill, or a delegated ingest objective) — never a build or export, and never a separate concept step. Do not offer "rebuild the concepts" alongside build/export as if it were the same family of action.',
     'Templates are instruction-only specs and deliverables are regenerated from them. When asked to change what a generated document says, edit the underlying wiki content (wiki_write_page) or the template\'s [[INSTRUCTION: ...]] sections — never write finished prose into a template, because a build copies it verbatim and it can no longer be refreshed from the wiki. template_write refuses prose outside an instruction block, so keep every sentence inside one.',
     'Promise only what the resolved capability actually exposes in its declared contract (the input schema the specialized agent publishes for that capability). When the user requests an execution parameter — a batch or chunk size, a count "N at a time", concurrency, ordering, priority, or any tuning knob — apply it only if that parameter exists in the target capability\'s published input schema. Otherwise do not confirm or promise it: delegate the objective, and if the user explicitly asked for that parameter, say plainly in one line that you started the work but do not control that aspect (the runtime and the specialized agent decide it). Never state or imply a parameter was applied when the agent contract cannot enforce it.',
     'If runtime__delegate returns a blocker or no specialized provider is available, report only that concrete blocker concisely. Never replace the missing execution path with a suggested slash command, skill, MCP tool name, manual file move, administrator escalation, or alternative workflow unless the user explicitly asks for alternatives.',
@@ -1266,7 +1268,7 @@ export function buildAgentSystemPrompt(state) {
       ? `Workspace profile (.wiki/profile.md) — durable user preferences, apply these to every reply (tone, tutoiement/vouvoiement, formatting, etc.):\n${workspaceProfile}`
       : null,
     currentArtifactPromptLine(currentArtifactFor(state.session)),
-    'Runtime control: you have runtime__status, runtime__cancel, runtime__kill and runtime__enqueue. When the user asks to stop, remove, clean or kill the current run, its jobs or the queue ("supprime le job et la queue", "arr\u00eate tout"), call runtime__kill (or runtime__cancel for a soft stop of just the run) and confirm what was stopped. When the user explicitly asks to delete, reset, abandon or replace the current plan, call runtime__kill with purge=true; never set purge=true for a simple stop. For questions about what is running or queued, call runtime__status and answer from its data. You have no approval tool: a pending approval is granted only by the user through the approval button or the /approve command. Never grant, claim or report an approval yourself; when the user asks to proceed with pending mutations, tell them to use those controls. When the user asks for a NEW action while a run is active, do not execute it: propose runtime__enqueue (run it after) or, if they insist it replaces the current work, runtime__kill then the new action.',
+    'Runtime control: you have runtime__status, runtime__cancel, runtime__kill and runtime__enqueue. When the user asks to stop, remove, clean or kill the current run, its jobs or the queue ("supprime le job et la queue", "arr\u00eate tout"), call runtime__kill (or runtime__cancel for a soft stop of just the run) and confirm what was stopped. When the user explicitly asks to delete, reset, abandon or replace the current plan, call runtime__kill with purge=true; never set purge=true for a simple stop. For questions about what is running or queued, call runtime__status and answer from its data. You have no approval tool: a pending approval is granted only by the user through the approval button or the /approve command. Never grant, claim or report an approval yourself; when the user asks to proceed with pending mutations, tell them to use those controls. A request to approve, validate, confirm or accept a pending run is a human control action: NEVER call runtime__delegate (or any capability) for it — answer with the control to use and nothing else. When the user asks for a NEW action while a run is active, do not execute it: propose runtime__enqueue (run it after) or, if they insist it replaces the current work, runtime__kill then the new action.',
     'When the user asks to refresh, show, or update the displayed plan or status, call runtime__status. This is a state refresh request, not a new business capability, and must never be delegated.',
     'Report every runtime control outcome exactly as the tool returned it \u2014 never embellish. If runtime__kill reports 0 run(s)/0 task(s)/0 purged, say there was nothing active to stop or purge; do NOT claim a run, plan, pending approval or queue item was removed. If runtime__status returns an error or could not be read, say the runtime state could not be retrieved and do not describe a state you never obtained. Never assert that something was cleaned, cancelled, approved or purged unless that specific tool result confirms it.',
     'Durable profile updates are actions in this stabilized version: delegate them instead of writing directly.',
@@ -1751,7 +1753,7 @@ export function createAgentGraph(options = {}) {
             forceDelegation: canDelegate,
           };
         }
-        const failure = 'Réponse rejetée : Donna a exposé une instruction interne ou une procédure manuelle incorrecte.';
+        const failure = 'Response rejected: Donna exposed an internal instruction or an incorrect manual procedure.';
         emitAgentEvent(state.session, 'assistant_message', 'agent_guard', { content: failure });
         return { response: failure, pendingToolCalls: null, readyToStream: false };
       }

package/src/agent/skillRecursion.test.js

@@ -98,21 +98,22 @@ test('borne la profondeur même sans cycle', async () => {
 });
 
 /*
- Cascade observée sur un `/wiki-ingest` : deux `/wiki-rebuild-concepts`, puis
- `/wiki-reclassify`, puis `/wiki-taxonomy`. Aucun cycle — trois compétences
- distinctes — mais la grille de concepts et la taxonomie reconstruites
- plusieurs fois pour une seule demande. La deuxième intention compilée de
- `/wiki-ingest` est mot pour mot le corps de `/wiki-rebuild-concepts` : le
- sélecteur par description la reconnaissait légitimement.
+ Cascade observée à l'époque des compétences concepts : un `/wiki-ingest`
+ relançait les compétences voisines plusieurs fois pour une seule demande,
+ parce que la deuxième intention compilée était mot pour mot le corps d'une
+ compétence sœur et que le sélecteur par description la reconnaissait
+ légitimement. Ces compétences (rebuild-concepts/reclassify/taxonomy) ont
+ disparu avec la simplification 0.15.66 ; la garde contre la récursion, elle,
+ reste — c'est elle que ces tests verrouillent.
 */
 const callWithObjective = (state, skillName, objective) =>
   handleRuntimeControlTool(state, 'run_skill', { skillName, _userInput: objective })
     .then((raw) => JSON.parse(raw));
 
 test('refuse une compétence voisine que l’intention décrit sans la nommer', async () => {
-  const objective = 'Run the production pipeline steps concepts, reclassify-concepts and taxonomy, in that order.';
+  const objective = 'Run the production pipeline steps ingest, build, export and polish, in that order.';
   const ran = [];
-  const result = await callWithObjective(session(['wiki-ingest'], ran), 'wiki-rebuild-concepts', objective);
+  const result = await callWithObjective(session(['wiki-ingest'], ran), 'wiki-build', objective);
 
   assert.equal(result.ok, false);
   assert.equal(result.code, 'nested_skill_match_blocked');
@@ -145,20 +146,20 @@ test('un chemin de fichier commençant par un nom de compétence ne vaut pas inv
 });
 
 test('hors de toute chaîne, la sélection par description reste permise', async () => {
-  const result = await callWithObjective(session(undefined), 'wiki-taxonomy', 'republish the graph taxonomy');
+  const result = await callWithObjective(session(undefined), 'wiki-build', 'regenerate the deliverables from the templates');
 
   assert.equal(result.ok, true);
 });
 
 /*
  Le nom seul ne prouve rien. Plusieurs compétences du scaffold portent un nom
- qui est aussi un mot courant : « Run the production pipeline steps concepts,
- reclassify-concepts and taxonomy » nomme `pipeline`, dont le lancement rejoue
+ qui est aussi un mot courant : « Run the production pipeline steps ingest,
+ build, export and polish » nomme `pipeline`, dont le lancement rejoue
  ingest + build + export + polish. Une intention doit citer sa cible EN TANT QUE
  compétence, pas l'employer comme mot.
 */
 test('un nom employé comme mot courant ne vaut pas invocation', async () => {
-  const objective = 'Run the production pipeline steps concepts, reclassify-concepts and taxonomy, in that order.';
+  const objective = 'Run the production pipeline steps ingest, build, export and polish, in that order.';
   const ran = [];
   const result = await callWithObjective(session(['wiki-ingest'], ran), 'pipeline', objective);