@dotdrelle/wiki-manager 0.15.93 → 0.15.96

package/README.md

@@ -121,47 +121,60 @@ in isolated workspaces.
 
 ## How wikiLLM compares
 
-Several open projects now build a Markdown wiki with an LLM. They target
+Several open projects now build an LLM-maintained knowledge layer over your
+sources — some as a wiki, some as a retrieval and answer platform. 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.
+and who operates it*. Snapshot as of **September 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),
+Projects compared: [OpenWiki](https://github.com/langchain-ai/openwiki) and
+[DeepWiki-Open](https://github.com/asyncfuncai/deepwiki-open) (code / knowledge
+wikis), [RAGFlow](https://github.com/infiniflow/ragflow) and
+[Onyx](https://github.com/onyx-dot-app/onyx) (RAG platforms), and
 [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>
+<tr>
+<th><small>Need</small></th>
+<th><small><strong>wikiLLM</strong></small></th>
+<th><small><strong>Code / knowledge wikis</strong><br>OpenWiki · DeepWiki-Open</small></th>
+<th><small><strong>RAG platforms</strong><br>RAGFlow · Onyx</small></th>
+<th><small><strong>Graph pipeline</strong><br>GraphRAG</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>
+<tr><td><small>Input</small></td><td><small>✅ Business docs — Confluence, PDF, Office, SaaS</small></td><td><small>✅ Code repo; OpenWiki also personal sources (Notion, Gmail, Slack…)</small></td><td><small>✅ PDF/Office/images, Confluence, S3, 50+ connectors</small></td><td><small>🟡 Plain text</small></td></tr>
+<tr><td><small>Output</small></td><td><small>✅ Maintained wiki <strong>+ deliverables</strong></small></td><td><small>✅ Linked Markdown wiki (+ graph / diagrams)</small></td><td><small>🟡 Grounded RAG answers, not a wiki</small></td><td><small>❌ Entity graph + summaries</small></td></tr>
+<tr><td><small>Agents &amp; actions</small></td><td><small>✅ Capability dispatcher, external agentic runtime, MCP tools</small></td><td><small>✅ Doc-generation agents, MCP connectors</small></td><td><small>✅ Agentic workflow, MCP actions, deep research</small></td><td><small>❌ Indexing pipeline</small></td></tr>
+<tr><td><small>Keep current</small></td><td><small>✅ Re-ingest / scheduled</small></td><td><small>✅ <code>--update</code> / regenerate, CI</small></td><td><small>✅ Scheduled connector sync</small></td><td><small>🟡 Maintenance mode</small></td></tr>
+<tr><td><small>Team UI</small></td><td><small>✅ Web console — wiki, graph, chat, runs</small></td><td><small>🟡 CLI + visualizer; DeepWiki-Open web app</small></td><td><small>✅ Self-hosted web app</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 repo / directory</small></td><td><small>🟡 Datasets / teams, not isolated services</small></td><td><small>❌ One index per corpus</small></td></tr>
+<tr><td><small>Human approval &amp; governance</small></td><td><small>✅ Approval-gated dispatcher, budgets, crash recovery</small></td><td><small>❌ PR review only</small></td><td><small>🟡 RBAC / SSO (Onyx enterprise)</small></td><td><small>❌</small></td></tr>
+<tr><td><small>License</small></td><td><small>❌ PolyForm <strong>Noncommercial</strong></small></td><td><small>✅ MIT</small></td><td><small>✅ Apache-2.0 / MIT CE</small></td><td><small>✅ MIT</small></td></tr>
 </tbody>
 </table>
 
+<small>Per-project details: OpenWiki emits OKF v0.2 and maintains a versioned
+evidence graph of “grounded claims”; GraphRAG is a research project now in
+maintenance mode; Onyx’s SSO/RBAC live in its enterprise edition.</small>
+
 **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.
+- **OpenWiki** and **DeepWiki-Open** build wikis from a **code repository**
+  (OpenWiki also ingests personal sources — Notion, Gmail, Slack…). Their input
+  is code or personal feeds, not a stack of business documents feeding your
+  deliverable templates.
+- **RAGFlow** and **Onyx** are RAG platforms: they index sources and answer
+  questions from them, with agents and connectors, but the output is an answer,
+  not a wiki you browse or a document you ship.
+- **GraphRAG** builds retrieval structure and is now in maintenance mode — a
+  complementary back end, not a wiki.
+- **wikiLLM** is the only one 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):**
 
@@ -170,8 +183,6 @@ Projects compared: [OpenWiki](https://github.com/langchain-ai/openwiki),
 - 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).
-- 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.
 

package/docker-compose.yml

@@ -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=${PRODUCTION_ALLOWED_STEPS:-doctor,doctor_apply,ingest,ingest_plan,ingest_apply,build,export,polish,restore,pipeline}
+      - PRODUCTION_ALLOWED_STEPS=${PRODUCTION_ALLOWED_STEPS:-doctor,doctor_apply,copy,ingest,ingest_plan,ingest_apply,ingest_rebuild,build,export,polish,restore,pipeline,lint}
       - 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": "0.15.93",
+  "version": "0.15.96",
   "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/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/wikiWorkspaceStart.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",
+    "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/wikiWorkspaceStart.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/deltaCoalescer.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 src/runtime/totp.test.js src/runtime/loginSession.test.js src/runtime/loginRoutes.test.js",
     "check-versions": "node scripts/check-versions.js",
     "prepack": "node scripts/check-versions.js",
     "prepublishOnly": "node scripts/check-versions.js",

package/src/agent/graph.js

@@ -31,6 +31,7 @@ import { artifactFromToolCall, currentArtifactFor, currentArtifactPromptLine, re
 import { capabilityRegistryForSession } from '../orchestrator/capabilityRegistry.js';
 import { objectiveForResolution } from '../orchestrator/objectiveResolver.js';
 import { fetchRuntimeState, postRuntimeCancel, postRuntimeControl, postRuntimeDelegate, postRuntimeKill, postRuntimeSkill } from '../runtime/client.js';
+import { formatPublicSkillInvocation, generateSkillAcknowledgment } from '../runtime/skillRun.js';
 
 const MAX_TOOL_ITERATIONS = 80;
 /**
@@ -555,6 +556,27 @@ function isUnresolvedTargetFailure(rawFailure) {
   return /file does not exist|does not exist|no files match/i.test(rawFailure);
 }
 
+// An agent that plans zero tasks is not a failure: it found nothing to act on
+// (no pending source, no template, no deliverable, no archived source to
+// re-file) and says so in its own synthesis sentence. Relaying that outcome is
+// the whole point — the old generic path called it "could not be started" and
+// sent the reader hunting a connectivity problem that did not exist.
+export function nothingToDoForDonna(rawReason) {
+  const reason = String(rawReason ?? '')
+    .replace(/\s*Available capabilities:\s*[\s\S]*$/i, '')
+    .replace(/\b(?:provider|endpoint)=[^\s,]+/g, '')
+    .replace(/\s{2,}/g, ' ')
+    .trim()
+    .slice(0, 400);
+  return JSON.stringify({
+    delegated: false,
+    nothingToDo: true,
+    reason,
+    instruction:
+      'The agent ran its planning and found nothing to act on. Answer the user naturally in their language, stating that plainly (use the reason above), and do not call it a failure, do not suggest restarting a service, and do not retry the same action. Do not expose exception names, capability identifiers, tool names, UUIDs, or internal routing details.',
+  });
+}
+
 function unresolvedTargetForDonna(rawFailure) {
   const cleaned = String(rawFailure ?? '')
     .replace(/\b(?:provider|endpoint)=[^\s,]+/g, '')
@@ -941,14 +963,19 @@ export async function handleRuntimeControlTool(session, tool, args = {}) {
         });
       }
       const result = await postRuntimeDelegate(objective, { url, workspace });
-      return result?.runId
-        ? JSON.stringify({
-            delegated: true,
-            runId: result.runId,
-            summary: result.delegation ?? null,
-            message: `Action started (${String(result.runId).slice(0, 8)}) after real-plan validation: ${result.delegation?.tasks ?? 0} task(s), ${result.delegation?.agent ?? 'resolved agent'}. Execution in progress.`,
-          })
-        : `Delegation refused: ${result?.error ?? JSON.stringify(result)}`;
+      if (result?.runId) {
+        return JSON.stringify({
+          delegated: true,
+          runId: result.runId,
+          summary: result.delegation ?? null,
+          message: `Action started (${String(result.runId).slice(0, 8)}) after real-plan validation: ${result.delegation?.tasks ?? 0} task(s), ${result.delegation?.agent ?? 'resolved agent'}. Execution in progress.`,
+        });
+      }
+      const refused = String(result?.error ?? JSON.stringify(result));
+      // Same EMPTY_PLAN outcome as the in-run path above: relay it, don't call
+      // it a failure.
+      if (/^EMPTY_PLAN:\s*/i.test(refused)) return nothingToDoForDonna(refused.replace(/^EMPTY_PLAN:\s*/i, ''));
+      return `Delegation refused: ${refused}`;
     }
     if (tool === 'run_skill') {
       const skillName = String(args.skillName ?? '').trim();
@@ -1351,24 +1378,6 @@ export function buildAgentSystemPrompt(state) {
   return customPrompt ? `${customPrompt}\n\n${agentContext}` : agentContext;
 }
 
-export function buildLimitedAgentResponse(state, reason = 'no workspace loaded with .wikirc.yaml') {
-  const workspace = state.session.workspace ?? 'no workspace selected';
-  const wikirc = state.session.wikirc?.profile ?? 'no profile loaded';
-  return [
-    `Donna is active. Current workspace: ${workspace}.`,
-    `Current wikirc profile: ${wikirc}.`,
-    '',
-    'I am the shell agent mode: use `/agent` to route free text through this LangGraph graph, or `/chat` for direct chat.',
-    `LLM connection: limited mode (${reason}).`,
-    `Available primitives: ${commandList(state.session)}.`,
-    '',
-    'Connected MCP tools:',
-    formatMcpToolsForAgent(state.session.mcp),
-    '',
-    'Limited mode: workspace, Docker Compose tools, MCP calls, /wiki fallback, skill discovery, and headless mode are wired.',
-    'Use `/help` to see deterministic shell commands.',
-  ].join('\n');
-}
 
 export function formatLlmUnavailableMessage(reason) {
   const clean = String(reason ?? 'unknown reason').replace(/\s+/g, ' ').trim();
@@ -1877,6 +1886,7 @@ export function createAgentGraph(options = {}) {
     const toolCalls = state.pendingToolCalls ?? [];
     const toolResultMessages = [];
     let terminalFailure = null;
+    let skillLaunch = null;
 
     for (const call of toolCalls) {
       const resolved = resolveToolCallName(state.session.mcp, call.function.name, INTERNAL_TOOL_SERVERS);
@@ -1991,6 +2001,19 @@ export function createAgentGraph(options = {}) {
             if (skillResult?.terminal === true) {
               terminalFailure = skillResult.code ?? 'skill_failed';
               ok = false;
+            } else if (skillResult?.accepted === true) {
+              // The skill owns execution from here. Without this the turn kept
+              // going and the model could re-delegate or refuse the very
+              // objective it had just launched — the "launched, then no agent
+              // capable" contradiction. End the turn with the launch
+              // acknowledgement instead of another tool round.
+              skillLaunch = {
+                publicInput: formatPublicSkillInvocation(
+                  skillResult.skill ?? args.skillName ?? '',
+                  args.arguments && typeof args.arguments === 'object' ? args.arguments : {},
+                ),
+                objectives: Number(skillResult.objectiveCount ?? skillResult.objectives ?? 1) || 1,
+              };
             }
           }
           if (tool === 'delegate' && /^Runtime control error \(delegate\):/i.test(resultText)) {
@@ -2009,6 +2032,14 @@ export function createAgentGraph(options = {}) {
                 missingRequiredFields: needsInput[1].split(',').map((item) => item.trim()).filter(Boolean),
                 instruction: 'Ask the user for the missing required information. Do not expose internal validation details.',
               });
+            } else if (/^EMPTY_PLAN:\s*/i.test(delegationFailure)) {
+              // Valid planning, nothing to do (no pending source/template/
+              // deliverable). Not a failure and not retryable — Donna relays
+              // the agent's own sentence instead of a connectivity-sounding
+              // generic error. Checked before the target heuristic: a plan
+              // reason is authoritative and could mention a missing file
+              // without meaning "unresolved target".
+              resultText = nothingToDoForDonna(delegationFailure.replace(/^EMPTY_PLAN:\s*/i, ''));
             } else if (isUnresolvedTargetFailure(delegationFailure)) {
               // A named target that resolves to nothing is not an unsupported
               // action: Donna can look it up and retry (or ask), so the turn
@@ -2147,6 +2178,24 @@ export function createAgentGraph(options = {}) {
         invalidResponseRetries: 0,
       };
     }
+    if (skillLaunch) {
+      // Deterministic, localized acknowledgement, generated in the session
+      // language like the `/turn` skill path. No further model turn: the skill
+      // is launched and owns execution, so nothing can contradict it.
+      const response = await generateSkillAcknowledgment(state.session, skillLaunch).catch(
+        () => `Started ${skillLaunch.publicInput} — ${skillLaunch.objectives} step(s) in progress.`,
+      );
+      return {
+        messages: toolResultMessages,
+        pendingToolCalls: null,
+        response,
+        forceDelegation: false,
+        retryWithoutTool: false,
+        terminalToolFailure: false,
+        invalidToolCallRetries: 0,
+        invalidResponseRetries: 0,
+      };
+    }
     return {
       messages: toolResultMessages,
       pendingToolCalls: null,
@@ -2159,6 +2208,9 @@ export function createAgentGraph(options = {}) {
   }
 
   function routeToolExecutor(state) {
+    // A skill launch already produced its final acknowledgement: end the turn
+    // instead of routing back for another model call that could contradict it.
+    if (state.response != null) return END;
     return 'orchestrator';
   }
 

package/src/agent/graph.test.js

@@ -3,7 +3,7 @@ import test from 'node:test';
 import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
-import { bareToolCallJson, buildAgentSystemPrompt, connectorConfigurationTarget, createAgentGraph, invalidSuggestedSlashCommands, invalidUserFacingToolNames, isOrchestrationBypassTool, knownCapabilityIds, normalizeToolArgumentsFromSchema } from './graph.js';
+import { bareToolCallJson, buildAgentSystemPrompt, connectorConfigurationTarget, createAgentGraph, invalidSuggestedSlashCommands, invalidUserFacingToolNames, isOrchestrationBypassTool, knownCapabilityIds, normalizeToolArgumentsFromSchema, nothingToDoForDonna } from './graph.js';
 
 test('user-facing response guard hides MCP identifiers generically', () => {
   const session = sessionBase();
@@ -13,6 +13,29 @@ test('user-facing response guard hides MCP identifiers generically', () => {
   );
 });
 
+test('an empty agent plan is relayed as "nothing to do", not as a failure', () => {
+  // The agent planned zero tasks and said why: no pending source, no template,
+  // no deliverable. That is an outcome. The old generic path called it "could
+  // not be started" and sent the reader after a connectivity problem that did
+  // not exist.
+  const payload = JSON.parse(nothingToDoForDonna(
+    'No Markdown files were found in raw/untracked; no ingest operation was planned.',
+  ));
+  assert.equal(payload.delegated, false);
+  assert.equal(payload.nothingToDo, true);
+  assert.match(payload.reason, /No Markdown files were found in raw\/untracked/);
+  assert.match(payload.instruction, /do not call it a failure/i);
+  assert.doesNotMatch(payload.instruction, /EMPTY_PLAN/);
+});
+
+test('the nothing-to-do reason strips internal routing details', () => {
+  const payload = JSON.parse(nothingToDoForDonna(
+    'No templates were found. Available capabilities: document.build, knowledge.update provider=production endpoint=http://host:3202',
+  ));
+  assert.doesNotMatch(payload.reason, /Available capabilities/);
+  assert.doesNotMatch(payload.reason, /provider=|endpoint=/);
+});
+
 test('CME setup stays direct while CME export execution stays orchestrated', () => {
   assert.equal(isOrchestrationBypassTool('cme__cme_export_run'), true);
   assert.equal(isOrchestrationBypassTool('cme__cme_setup'), false);
@@ -601,7 +624,10 @@ test('an explicitly selected skill runs through the intra-runtime path with name
     },
   });
   const result = await createAgentGraph().invoke({ input: 'lance le skill deliver avec le template Quarterly report', session });
-  assert.equal(result.response, 'Skill mis en file.');
+  // Launching a skill ends the turn: the acknowledgement is generated once and
+  // the model is not given a second chance to re-delegate or contradict it.
+  assert.equal(result.response, 'Started /deliver deliverable="Quarterly report" — 1 step(s) in progress.');
+  assert.equal(mainCalls, 1);
   // `skillStack` accompagne désormais la demande : le run imbriqué démarre après
   // le nettoyage de celui-ci, et c'est le seul canal par lequel il peut savoir
   // quelles compétences sont déjà ouvertes au-dessus de lui.

package/src/cli/wiki-manager.js

@@ -11,6 +11,7 @@ import { randomUUID } from 'node:crypto';
 import { spawnSync } from 'node:child_process';
 import { readFileSync } from 'node:fs';
 import { mkdir, writeFile } from 'node:fs/promises';
+import { createInterface } from 'node:readline/promises';
 import { dirname, join, resolve } from 'node:path';
 import { fileURLToPath } from 'node:url';
 import { ensureManagerScaffold, loadManagerEnv } from '../core/env.js';
@@ -28,6 +29,8 @@ import { extractActivity, mergePolledActivity, parseJsonText, sessionActivities,
 import { syncActivitiesToPlan, formatPlanStatus } from '../core/plan.js';
 import { createAgentEvent, dispatchAgentEvent, reduceAgentEvents } from '../core/agentEvents.js';
 import { runAgentTurn, runAgenticLoop } from '../core/agentLoop.js';
+import { createDeltaCoalescer } from '../runtime/deltaCoalescer.js';
+import { TERMINAL_STATUS_SET } from '../orchestrator/taskStatuses.js';
 import { resolveCapabilityConcurrency } from '../orchestrator/scheduler.js';
 import { capabilityRegistryForSession } from '../orchestrator/capabilityRegistry.js';
 import { CapabilityUnavailableError, resolve as resolveCapability } from '../orchestrator/capabilityResolver.js';
@@ -49,6 +52,17 @@ function valueAfter(argv, flag) {
   return argv[index + 1];
 }
 
+async function confirmOnTty(question, defaultValue) {
+  if (!process.stdin.isTTY || !process.stdout.isTTY) return defaultValue;
+  const readline = createInterface({ input: process.stdin, output: process.stdout });
+  try {
+    const answer = (await readline.question(`${question} [y/N] `)).trim().toLowerCase();
+    return answer === 'y' || answer === 'yes';
+  } finally {
+    readline.close();
+  }
+}
+
 function errorDiagnostic(err) {
   const parts = [];
   let current = err;
@@ -815,7 +829,7 @@ async function runHeadless(argv, agent) {
 // export finished, before the ingest had even started. The control queue is the
 // only place where the whole chain is observable, so the wait is scoped to
 // chainId and ends when every item of that chain is terminal.
-const CHAIN_TERMINAL_STATUSES = new Set(['done', 'failed', 'cancelled', 'skipped']);
+const CHAIN_TERMINAL_STATUSES = TERMINAL_STATUS_SET;
 
 export async function waitForRuntimeChain(session, log, {
   chainId,
@@ -949,6 +963,11 @@ async function runRuntime(argv, agent) {
   const host = valueAfter(argv, '--host') ?? process.env.WIKI_MANAGER_RUNTIME_HOST ?? '0.0.0.0';
   const port = Number(valueAfter(argv, '--port') ?? process.env.WIKI_MANAGER_RUNTIME_PORT ?? 7788);
   const stateDir = valueAfter(argv, '--state-dir') ?? defaultRuntimeStateDir();
+  // The TOTP login modules resolve their state directory from the ENV, not
+  // from the CLI option: without this, `runtime --state-dir X` would store
+  // totp.json/session.json in the DEFAULT manager state dir while the rest of
+  // the runtime uses X — sessions issued on one side invisible to the other.
+  process.env.WIKI_MANAGER_STATE_DIR = stateDir;
   const auth = resolveRuntimeAuthToken({ host, stateDir });
   if (auth.token) process.env.WIKI_MANAGER_RUNTIME_TOKEN = auth.token;
   if (!Number.isInteger(port) || port <= 0 || port > 65535) {
@@ -1351,7 +1370,15 @@ async function runRuntime(argv, agent) {
       );
     }
     if (!Array.isArray(fragment?.tasks) || fragment.tasks.length === 0) {
-      throw new Error(fragment?.summary?.initialSynthesis?.[0] ?? `No task was planned for ${selection.capability}/${selection.operation}.`);
+      // An agent may legitimately plan nothing: no pending source to ingest, no
+      // template, no deliverable, no archived source to re-file. That is an
+      // OUTCOME, not a failure — the agent carries its own human-readable
+      // sentence in summary.initialSynthesis. The stable EMPTY_PLAN sentinel
+      // tells graph.js to relay that outcome instead of the generic "could not
+      // be started" failure (which read as a connectivity problem).
+      const plannedReason = String(fragment?.summary?.initialSynthesis?.[0] ?? '').replace(/\s+/g, ' ').trim();
+      const reason = plannedReason || `${selection.capability} planned no task for ${selection.operation}.`;
+      throw new Error(`EMPTY_PLAN: ${reason}`);
     }
     const validation = validateFragment(fragment, {
       registry: capabilityRegistryForSession(session),
@@ -1714,7 +1741,21 @@ async function runRuntime(argv, agent) {
       workspace: context.workspace ?? ephemeral.workspace ?? null,
     }));
     const messages = conversationSeed({ agentProjection: persistedProjection }, input);
+    // Streaming fragments are coalesced before they are persisted and pushed —
+    // one synchronous SQLite insert (plus one SSE write) per token stalled the
+    // event loop, freezing both chats (serve and ShellUI) while a long answer
+    // was still being produced. Flushed before any non-delta event so ordering
+    // and the final tail are preserved.
+    const deltaCoalescer = createDeltaCoalescer((delta) => dispatchAgentEvent(ephemeral, createAgentEvent('assistant_delta', {
+      origin: 'runtime_turn',
+      turnId,
+      workspace: context.workspace ?? null,
+      payload: { delta },
+    })), { intervalMs: 80 });
     ephemeral._onAgentEvent = (event) => {
+      // Never let a final message (or any other event) overtake the buffered
+      // fragments that precede it: flush them first, in order.
+      if (event.type !== 'assistant_delta' && event.type !== 'assistant_delta_reset') deltaCoalescer.flush();
       const interactiveEvent = {
         ...event,
         origin: 'runtime_turn',
@@ -1760,28 +1801,30 @@ async function runRuntime(argv, agent) {
       response = await runHeadlessChatTurn(ephemeral, input, {
         history,
         onStep: ephemeral._onStep,
-        // Fragments de réponse publiés au fil de l'eau. Le réducteur les
-        // agrège dans la dernière entrée de conversation (`assistant_delta`),
-        // que `assistant_message` vient ensuite figer : les deux interfaces
-        // voient la réponse s'écrire, au lieu d'attendre le tour complet.
-        onTextDelta: (delta) => dispatchAgentEvent(ephemeral, createAgentEvent('assistant_delta', {
-          origin: 'runtime_turn',
-          turnId,
-          workspace: context.workspace ?? null,
-          payload: { delta },
-        })),
-        onTextReset: () => dispatchAgentEvent(ephemeral, createAgentEvent('assistant_delta_reset', {
-          origin: 'runtime_turn',
-          turnId,
-          workspace: context.workspace ?? null,
-          payload: {},
-        })),
+        // Fragments de réponse publiés au fil de l'eau, coalescés (voir
+        // deltaCoalescer ci-dessus). Le réducteur les agrège dans la dernière
+        // entrée de conversation (`assistant_delta`), que `assistant_message`
+        // vient ensuite figer : les deux interfaces voient la réponse s'écrire
+        // sans qu'un insert SQLite par token ne bloque le flux.
+        onTextDelta: (delta) => deltaCoalescer.push(delta),
+        onTextReset: () => {
+          deltaCoalescer.reset();
+          dispatchAgentEvent(ephemeral, createAgentEvent('assistant_delta_reset', {
+            origin: 'runtime_turn',
+            turnId,
+            workspace: context.workspace ?? null,
+            payload: {},
+          }));
+        },
         openWikiPages,
       });
     } else {
       ephemeral.openWikiPages = openWikiPages;
       response = await runAgentTurn(agent, ephemeral, input, { messages, signal });
     }
+    // Flush the tail before the turn is finalized, then stop the timer.
+    deltaCoalescer.flush();
+    deltaCoalescer.dispose();
     // Persist the artifact the turn may have opened/edited (template_write,
     // template_read, …) back onto the long-lived session, so the next /turn —
     // chat or agent — sees it. The ephemeral session is otherwise discarded.
@@ -1898,6 +1941,65 @@ export async function runCli(argv) {
     return;
   }
 
+  if (argv[0] === 'login' || argv[0] === 'logout') {
+    const { ensureRuntime } = await import('../runtime/lifecycle.js');
+    const { fetchLoginStatus, requestTotpSession, revokeRuntimeSession } = await import('../runtime/totpLogin.js');
+    let runtime = null;
+    try {
+      runtime = await ensureRuntime();
+    } catch (err) {
+      runtime = { url: null, error: err instanceof Error ? err.message : String(err) };
+    }
+    if (!runtime.url) throw new Error(`Runtime unavailable: ${runtime.error}`);
+    if (argv[0] === 'logout') {
+      const { currentSessionToken } = await import('../runtime/loginSession.js');
+      const revoked = await revokeRuntimeSession(runtime, currentSessionToken());
+      console.log(revoked ? 'Session revoked.' : 'No active session to revoke (or the runtime is not answering).');
+      return;
+    }
+    if (argv.includes('--reset')) {
+      // Re-enrollment after a lost authenticator: wipes the secret and the
+      // active session (local files, not an HTTP route) then runs the normal
+      // login flow, which lands on a fresh enrollment QR code.
+      const confirmed = argv.includes('--yes') || (await confirmOnTty(
+        'Reset the TOTP enrollment? This revokes the active session.\nContinue?',
+        false,
+      ));
+      if (!confirmed) {
+        console.log('Aborted.');
+        return;
+      }
+      const { resetTotpEnrollment } = await import('../runtime/loginSession.js');
+      resetTotpEnrollment();
+      console.log('TOTP enrollment and the active session were reset — enrolling a new authenticator now.');
+    }
+    const status = await fetchLoginStatus(runtime);
+    if (status && !status.enabled) {
+      console.log('TOTP login is disabled (WIKI_MANAGER_TOTP=off).');
+      return;
+    }
+    if (!argv.includes('--reset') && status?.sessionActive) {
+      // A session already exists, but the *browser* only gets the shared
+      // wiki_session cookie the moment it verifies a code on the runtime's
+      // origin. Re-open the page when asked so a browser that has none can
+      // seed it (and then reuse it on serve, same host) without waiting for
+      // the session to expire first.
+      if (!argv.includes('--no-open')) {
+        const { loginPageUrl, openBrowser } = await import('../runtime/totpLogin.js');
+        try {
+          await openBrowser(loginPageUrl(runtime));
+        } catch {
+          console.log(`Open ${loginPageUrl(runtime)} to seed this browser's session.`);
+        }
+      }
+      console.log(`Session already active until ${new Date(status.sessionExpiresAt).toLocaleString()}.`);
+      return;
+    }
+    const result = await requestTotpSession(runtime, { open: !argv.includes('--no-open') });
+    if (!result.ok) throw new Error(result.error);
+    return;
+  }
+
   if (argv.includes('--setup-wizard')) {
     if (!process.versions.bun) {
       throw new Error('Setup wizard requires Bun. Run: bun ./bin/wiki-manager.js --setup-wizard');
@@ -2008,6 +2110,15 @@ export async function runCli(argv) {
       runtime = unavailableRuntime(err);
       console.error(`Runtime unavailable: ${runtime.error}`);
     }
+    if (runtime.url) {
+      const { requestTotpSession } = await import('../runtime/totpLogin.js');
+      const gate = await requestTotpSession(runtime);
+      if (!gate.ok) {
+        console.error(gate.error);
+        process.exitCode = 1;
+        return;
+      }
+    }
     preflight = withRuntimePreflight(preflight, runtime);
     // render() resolves at mount; the TUI owns renderer teardown. The shared
     // runtime deliberately survives shell exit because `serve` may use it.
@@ -2030,6 +2141,15 @@ export async function runCli(argv) {
       runtime = unavailableRuntime(err);
       console.error(`Runtime unavailable: ${runtime.error}`);
     }
+    if (runtime.url) {
+      const { requestTotpSession } = await import('../runtime/totpLogin.js');
+      const gate = await requestTotpSession(runtime);
+      if (!gate.ok) {
+        console.error(gate.error);
+        process.exitCode = 1;
+        return;
+      }
+    }
   }
   await runShell({ agent, packageJson, runtime });
 }