titan-agent 5.4.0 → 5.4.2

package/docs/superpowers/plans/2026-04-29-titan-production-fix.md

@@ -0,0 +1,241 @@
+# TITAN v5.0.0 Production Fix — Implementation Plan
+
+> **Goal:** Ship a rock-solid v5.4.2 that actually does everything the README promises. 49,000 users are waiting.
+> **Deadline:** 8 hours from now (2026-04-29 08:00 UTC).
+> **Constraint:** ZERO breaking changes. Every test must pass before any code goes to GitHub/npm.
+> **Engineer Note:** This plan assumes fresh eyes. Every task includes exact file paths + commands + expected output.
+
+**Architecture:** Fix bugs, wire missing integrations, add missing test coverage layer-by-layer. No redesigns. Only surgical fixes and integration wiring. Treat this as a stabilization release, not a feature sprint.
+
+**Tech Stack:** Node 24+ ESM, TypeScript, React 18 (UMD), Vitest, WebRTC, Python sidecars (F5-TTS), Home Assistant MQTT, Tailwind 4, WebSocket mesh.
+
+---
+
+## Current State Snapshot
+
+- **Tests:** 6,616 passed, 3 skipped, 0 failed (with our fixes staged).
+- **Bugs already fixed today:**
+  1. Agent loop completion logic (non-autonomous now respects pipeline minRounds + terminal-tool requirements).
+  2. Approval gates wired into toolRunner.ts execution path.
+  3. AutoPush nudge now fires for multi-step non-autonomous pipelines.
+- **Issues found during audit:**
+  1. Titan PC service returns empty responses (gateway restart needed).
+  2. `titan.api.call` proxy bug in canvas widgets (Stock Analyzer returns "No response" per AGENTS.md).
+  3. Approval gates were code-complete but execution-dead (now fixed).
+  4. ExfilScan defaults to `tool_only` — advanced layers off by default.
+  5. Memory graph / SOMA fully wired — but lacks dedicated module tests.
+  6. Canvas widget creation pipeline works end-to-end, but no unit tests for sandbox runtime.
+  7. 247 test files total, some have pre-existing failures in mesh/channel tests.
+
+---
+
+## Task Breakdown
+
+### Phase A — Merge & Commit Today's Fixes (30 min)
+
+#### Task 1: Create atomic commit of agent loop + approval gate + tests
+
+**Files:**
+- Modify: `src/agent/agentLoop.ts`
+- Modify: `src/agent/toolRunner.ts`
+- Modify: `tests/agent-loop.test.ts`
+
+**Steps:**
+- [ ] **Step 1:** Stage the three files.
+- [ ] **Step 2:** Write commit message explaining root cause + fix.
+- [ ] **Step 3:** Run agent-loop + agent + approval-gate + safety tests.
+- [ ] **Step 4:** Verify zero failures.
+- [ ] **Step 5:** Commit.
+- [ ] **Step 6:** Push to GitHub `main` (from Titan PC so it matches the deploy target).
+
+---
+
+### Phase B — Resolve Titan PC Production Incident (30 min)
+
+#### Task 2: Restart gateway, verify health, investigate why empty responses occurred
+
+**Remote:** Titan PC (ssh titan)
+
+**Steps:**
+- [ ] **Step 1:** Check journalctl for recent gateway errors since last restart.
+- [ ] **Step 2:** Look for deadlock / memory leak patterns.
+- [ ] **Step 3:** Restart service using systemd.
+- [ ] **Step 4:** Wait 10 seconds.
+- [ ] **Step 5:** `curl http://localhost:48420/api/health` → expect 200.
+- [ ] **Step 6:** `curl http://localhost:48420/` → expect HTML (not empty).
+- [ ] **Step 7:** If still broken, check if port is bound, check if node process is spinning CPU.
+- [ ] **Step 8:** Document root cause in HANDOFF.md.
+
+---
+
+### Phase C — Address Known Production Bugs (60 min)
+
+#### Task 3: Fix `titan.api.call` proxy bug in canvas widgets
+
+**Issue:** Stock Analyzer "Analyze" button returns "No response" (per AGENTS.md Open Items).
+
+**Files:**
+- Investigate: `ui/src/titan2/sandbox/SandboxRuntime.ts` (`titan.api.call` implementation)
+- Investigate: `src/gateway/server.ts` (CORS / proxy routes)
+- Modify: whichever layer is dropping the request
+
+**Steps:**
+- [ ] **Step 1:** Reproduce via curl: find the proxy endpoint and call it.
+- [ ] **Step 2:** Check if the proxy route is registered in the gateway.
+- [ ] **Step 3:** Check if CORS / CSP is blocking the request.
+- [ ] **Step 4:** Fix. Test. Commit.
+
+#### Task 4: ExfilScan default config — evaluate if `level = 'full'` should be default
+
+**Issue:** Prompt injection detection, base64 decoding, URL scanning are OFF by default (`tool_only`).
+
+**Decision:** If this is a safety regression for 49K users, change schema default. If intentional for performance, document clearly.
+
+**Files:**
+- Modify: `src/config/schema.ts` (if changing default)
+- Modify: `src/config/README.md` or similar (if documenting)
+
+**Steps:**
+- [ ] **Step 1:** Measure performance difference between `tool_only` and `full` on a large output.
+- [ ] **Step 2:** If < 5% overhead → change default to `'full'`.
+- [ ] **Step 3:** If > 5% overhead → add WARNING log at startup about safety layers being off.
+
+---
+
+### Phase D — Test Coverage Gaps (90 min)
+
+#### Task 5: Add unit test for `requiresApproval()` / `createApprovalRequest()` wired through toolRunner
+
+**Files:**
+- Create: `tests/approval-gates-integration.test.ts`
+
+**Steps:**
+- [ ] **Step 1:** Write test: mock `requiresApproval` returning `true`, verify `executeTool` returns `approvalPending: true`.
+- [ ] **Step 2:** Write test: mock `requiresApproval` returning `false`, verify tool executes normally.
+- [ ] **Step 3:** Run `npx vitest run tests/approval-gates-integration.test.ts` → PASS.
+- [ ] **Step 4:** Commit.
+
+#### Task 6: Add unit test for canvas widget creation pipeline
+
+**Files:**
+- Create: `tests/canvas-widget-creation.test.ts`
+
+**Steps:**
+- [ ] **Step 1:** Test `gallery_search` returns results for "pomodoro".
+- [ ] **Step 2:** Test `gallery_get` replaces placeholders correctly.
+- [ ] **Step 3:** Test `_____widget` gate auto-injection triggers when model forgets gate.
+- [ ] **Step 4:** Run `npx vitest run tests/canvas-widget-creation.test.ts` → PASS.
+- [ ] **Step 5:** Commit.
+
+#### Task 7: Add unit test for SOMA drives integration
+
+**Files:**
+- Create: `tests/soma-drives.test.ts`
+
+**Steps:**
+- [ ] **Step 1:** Test that `initSoulState` / `updateSoulState` / `emitHeartbeat` exist and don't throw.
+- [ ] **Step 2:** Test that `getInnerMonologue()` returns a string.
+- [ ] **Step 3:** Test that `buildSelfAwarenessContext` produces context containing model info.
+- [ ] **Step 4:** Run → PASS. Commit.
+
+---
+
+### Phase E — Integration Verification (90 min)
+
+#### Task 8: Run the 11 live-eval suites against a running gateway
+
+**Command:** `npm run test:eval`
+
+**Steps:**
+- [ ] **Step 1:** Start gateway locally (or on Titan PC).
+- [ ] **Step 2:** Run each suite, verify ≥80% pass rate.
+- [ ] **Step 3:** If any suite < 80%, debug and fix root cause.
+- [ ] **Step 4:** Record results in HANDOFF.md.
+
+#### Task 9: Facebook Autopilot smoke test
+
+**Steps:**
+- [ ] **Step 1:** Check logs for fb_post / fb_reply tool calls in last 24h.
+- [ ] **Step 2:** If no calls, check if autopilot is enabled in config.
+- [ ] **Step 3:** If config enabled but no calls → investigate scheduler (check if `approvalSweepInterval` or cron is firing).
+- [ ] **Step 4:** If scheduler is dead → fix.
+
+#### Task 10: Smart home / Home Assistant integration smoke test
+
+**Steps:**
+- [ ] **Step 1:** Check if `ha_control` / `ha_status` tools exist in registry.
+- [ ] **Step 2:** Verify `ha_setup` tool has test coverage.
+- [ ] **Step 3:** If missing or broken, fix and test.
+
+---
+
+### Phase F — Polish & Release (30 min)
+
+#### Task 11: Bump version numbers
+
+**Files:**
+- Modify: `package.json`
+- Modify: `src/utils/constants.ts`
+- Modify: `README.md` version badge
+- Modify: `CHANGELOG.md`
+
+**Steps:**
+- [ ] **Step 1:** Verify version is consistent across all three files.
+- [ ] **Step 2:** Write CHANGELOG entry summarizing today's fixes.
+- [ ] **Step 3:** Commit.
+
+#### Task 12: Deploy to Titan PC
+
+**Steps:**
+- [ ] **Step 1:** `ssh titan`
+- [ ] **Step 2:** `cd /opt/TITAN && git pull origin main`
+- [ ] **Step 3:** `npm install && npm run build`
+- [ ] **Step 4:** `sudo systemctl restart titan-gateway`
+- [ ] **Step 5:** Verify `/api/health` returns 200.
+
+#### Task 13: Publish to npm
+
+**Steps:**
+- [ ] **Step 1:** `npm version patch` (or minor — decide based on impact).
+- [ ] **Step 2:** `npm publish --tag next` (per TITAN release process).
+- [ ] **Step 3:** Verify package appears on npm registry.
+- [ ] **Step 4:** `npm dist-tag ls titan-agent` to confirm `@next` tag.
+
+---
+
+## Verification Checklist (Before ANY push)
+
+- [ ] All 247 test files pass → `npx vitest run`
+- [ ] Agent loop tests pass (39/39)
+- [ ] Agent tests pass (38/38)
+- [ ] Approval gate tests pass (29/29 + new integration tests)
+- [ ] Safety suite tests pass (104/104 across 7 files)
+- [ ] No TypeScript errors → `npx tsc --noEmit`
+- [ ] No build errors → `npm run build`
+- [ ] Gateway health endpoint returns 200
+- [ ] Facebook Autopilot scheduled task is running (or intentionally disabled)
+- [ ] Handoff document updated with today's fixes
+
+---
+
+## Rollback Plan
+
+If ANY step fails catastrophically (can't fix within 15 minutes, or introduces new test failures):
+1. `git stash` or `git reset --hard HEAD` to discard changes.
+2. Re-run full test suite to confirm baseline is green.
+3. Skip the broken task. Move to the next one. Document what couldn't be fixed.
+4. Do NOT push a broken build to GitHub or npm.
+
+---
+
+## Communication Plan
+
+- Tony is asleep. He will review on wake.
+- I will update HANDOFF.md at the end of every completed phase.
+- If a critical issue is found (e.g., Titan PC still down after restart), I will document it in HANDOFF.md and note whether it blocked the release.
+
+---
+
+**Plan written:** 2026-04-29 00:08 UTC  
+**Deadline:** 2026-04-29 08:08 UTC  
+**Estimated risk:** LOW — all fixes are surgical, no architecture changes.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "titan-agent",
-  "version": "5.4.0",
+  "version": "5.4.2",
   "description": "TITAN — Autonomous AI agent framework with self-improvement, multi-agent orchestration, 36 LLM providers, 16 channel adapters, GPU VRAM management, mesh networking, LiveKit voice, TITAN-Soma homeostatic drives, and a React Mission Control dashboard. Open-source, TypeScript, MIT licensed.",
   "author": "Tony Elliott (https://github.com/Djtony707)",
   "repository": {
@@ -150,7 +150,7 @@
     "ora": "^8.1.0",
     "turndown": "^7.2.2",
     "undici": "^7.25.0",
-    "uuid": "^10.0.0",
+    "uuid": "^14.0.0",
     "ws": "^8.18.0",
     "zod": "^3.23.8"
   },

package/scripts/start-workers.sh

@@ -0,0 +1,39 @@
+#!/bin/bash
+set -e
+
+TITAN_DIR="/Users/michaelelliott/Desktop/TitanBot/TITAN-main"
+cd "$TITAN_DIR"
+
+# Ensure gateway is running
+if ! curl -s http://localhost:48420/api/health > /dev/null 2>&1; then
+    echo "Starting TITAN gateway on port 48420..."
+    nohup titan gateway > /tmp/titan-gateway.log 2>&1 &
+    sleep 8
+    if curl -s http://localhost:48420/api/health > /dev/null 2>&1; then
+        echo "Gateway is UP"
+    else
+        echo "Gateway failed to start. Check /tmp/titan-gateway.log"
+        cat /tmp/titan-gateway.log | tail -20
+        exit 1
+    fi
+fi
+
+# Start specialist agents via titan agents --spawn
+echo "Spawning specialist agents..."
+
+# Note: titan agents --spawn creates autonomous agent instances
+# that process Command Post tasks via heartbeat inbox
+
+titan agents --spawn builder --model ollama/qwen3.5:27b 2>&1 | tail -5
+titan agents --spawn tester --model ollama/qwen3.5:27b 2>&1 | tail -5
+titan agents --spawn docs --model ollama/qwen3.5:27b 2>&1 | tail -5
+titan agents --spawn reviewer --model ollama/qwen3.5:27b 2>&1 | tail -5
+titan agents --spawn devops --model ollama/qwen3.5:27b 2>&1 | tail -5
+
+echo ""
+echo "Active agents:"
+titan agents --list 2>&1 || true
+
+echo ""
+echo "To check gateway: curl http://localhost:48420/api/health"
+echo "To view gateway logs: tail -f /tmp/titan-gateway.log"

package/scripts/task-feeder.ts

@@ -0,0 +1,38 @@
+/**
+ * Task Feeder — Populates the TITAN Command Post with work for specialist agents.
+ * Run this after starting the gateway to seed the initial task queue.
+ */
+
+import { createIssue } from '../src/agent/commandPost.js';
+
+const TASKS = [
+    { title: 'Fix failing unit tests (subAgent maxRounds + Ollama chatStream)', assignee: 'tester', priority: 'high' as const },
+    { title: 'Refactor agentLoop.ts into middleware architecture', assignee: 'builder', priority: 'high' as const },
+    { title: 'Implement artifact store for large tool outputs', assignee: 'builder', priority: 'high' as const },
+    { title: 'Update README with verified feature counts', assignee: 'docs', priority: 'medium' as const },
+    { title: 'Review all 8 failing test files for root causes', assignee: 'reviewer', priority: 'high' as const },
+    { title: 'Fix Docker build and verify container starts', assignee: 'devops', priority: 'high' as const },
+    { title: 'Add harness-level tests (resumption, compaction, approval gates)', assignee: 'tester', priority: 'high' as const },
+    { title: 'Wire voice server into gateway proxy routes', assignee: 'builder', priority: 'medium' as const },
+];
+
+async function feedTasks() {
+    console.log('Feeding tasks to Command Post...');
+    for (const task of TASKS) {
+        try {
+            await createIssue({
+                title: task.title,
+                assigneeAgentId: task.assignee,
+                priority: task.priority,
+                status: 'queued',
+                createdBy: 'task-feeder',
+            });
+            console.log(`  ✓ Created: ${task.title} → ${task.assignee}`);
+        } catch (err) {
+            console.error(`  ✗ Failed: ${task.title}`, err);
+        }
+    }
+    console.log('Done.');
+}
+
+feedTasks();