@aion0/forge 0.5.21 → 0.5.23

package/lib/workspace/presets.ts

@@ -71,8 +71,9 @@ Rules:
 - Read docs/qa/ to see what was already tested. Skip tests that already passed for unchanged features.
 - Test plans are versioned: docs/qa/test-plan-v1.1.md
 - Test reports are versioned: docs/qa/test-report-v1.1.md
-- Test code goes in tests/ directory.
+- Test code goes in tests/e2e/ directory (Playwright tests).
 - Do NOT fix bugs — only report them clearly in your test report.
+- Use the run_plugin MCP tool to execute Playwright tests and take screenshots. Prefer run_plugin over raw bash commands.
 
 Communication rules:
 - Only send [SEND:...] messages for BLOCKING issues that prevent the product from working.
@@ -83,11 +84,12 @@ Communication rules:
     agentId: 'claude',
     dependsOn: [],
     workDir: './',
+    plugins: ['playwright'],
     outputs: ['tests/', 'docs/qa/'],
     steps: [
       { id: 'plan', label: 'Test Planning', prompt: 'Read the latest PRD in docs/prd/ and existing test plans in docs/qa/. Write a NEW test plan file covering only the NEW/CHANGED features.' },
-      { id: 'write-tests', label: 'Write Tests', prompt: 'Implement test cases in tests/ directory based on your test plan. Add new tests, do not rewrite existing passing tests. Do NOT send any messages to other agents in this step.' },
-      { id: 'execute', label: 'Execute Tests', prompt: 'Run all tests. Write a test report documenting results. Only if you find BLOCKING bugs (app crashes, data loss, security holes), send ONE consolidated message: [SEND:Engineer:fix_request] followed by a brief list of blocking issues. Minor issues go in the report only.' },
+      { id: 'write-tests', label: 'Write Tests', prompt: 'Write Playwright e2e test cases in tests/e2e/ directory based on your test plan. If playwright.config.ts does not exist, create one with testDir: "./tests/e2e" and baseURL from the project. Add new tests, do not rewrite existing passing tests. Do NOT send any messages to other agents in this step.' },
+      { id: 'execute', label: 'Execute Tests', prompt: 'Run tests using run_plugin (plugin: "playwright", action: "test"). If run_plugin is not available, fall back to: npx playwright test tests/e2e/ --reporter=line. Write a test report documenting results. Only if you find BLOCKING bugs (app crashes, data loss, security holes), send ONE consolidated message: [SEND:Engineer:fix_request] followed by a brief list of blocking issues. Minor issues go in the report only.' },
     ],
   },
 

package/lib/workspace/session-monitor.ts

@@ -29,7 +29,7 @@ export interface SessionMonitorEvent {
 
 const POLL_INTERVAL = 3000;      // check every 3s
 const IDLE_THRESHOLD = 3540000;  // 59min of no file change → check for result entry
-const STABLE_THRESHOLD = 3600000; // 60min of no change → force done (fallback if hook missed)
+const STABLE_THRESHOLD = 3600000; // 60min of no change → force done (fallback if hook missed) [59min check + 1min grace]
 
 export class SessionFileMonitor extends EventEmitter {
   private timers = new Map<string, NodeJS.Timeout>();
@@ -155,14 +155,14 @@ export class SessionFileMonitor extends EventEmitter {
       if (prevState === 'running') {
         if (stableFor >= IDLE_THRESHOLD) {
           // Check if session file has a 'result' entry at the end
-          const resultInfo = this.checkForResult(filePath);
+          const resultInfo = this.checkForResult(filePath, size);
           if (resultInfo) {
             this.setState(agentId, 'done', filePath, resultInfo);
             return;
           }
         }
         if (stableFor >= STABLE_THRESHOLD) {
-          // Force done after 30s even without result entry
+          // Force done after 60min even without result entry (ultimate fallback)
           this.setState(agentId, 'done', filePath, 'stable timeout');
           return;
         }
@@ -179,15 +179,19 @@ export class SessionFileMonitor extends EventEmitter {
    * Check the last few lines of the session file for a 'result' type entry.
    * Claude Code writes this when a turn completes.
    */
-  private checkForResult(filePath: string): string | null {
+  private checkForResult(filePath: string, fileSize?: number): string | null {
     try {
-      // Read last 4KB of the file
-      const stat = statSync(filePath);
-      const readSize = Math.min(4096, stat.size);
-      const fd = require('node:fs').openSync(filePath, 'r');
+      // Read last 4KB of the file — fileSize passed from caller to avoid second statSync
+      const { openSync, readSync, closeSync } = require('node:fs') as typeof import('node:fs');
+      const size = fileSize ?? statSync(filePath).size;
+      const readSize = Math.min(4096, size);
+      const fd = openSync(filePath, 'r');
       const buf = Buffer.alloc(readSize);
-      require('node:fs').readSync(fd, buf, 0, readSize, Math.max(0, stat.size - readSize));
-      require('node:fs').closeSync(fd);
+      try {
+        readSync(fd, buf, 0, readSize, Math.max(0, size - readSize));
+      } finally {
+        closeSync(fd);
+      }
 
       const tail = buf.toString('utf-8');
       const lines = tail.split('\n').filter(l => l.trim());

package/lib/workspace/types.ts

@@ -40,6 +40,8 @@ export interface WorkspaceAgentConfig {
   boundSessionId?: string;
   // Skip dangerous permissions check (default true when persistentSession is enabled)
   skipPermissions?: boolean;
+  // Plugins: list of plugin IDs (or source IDs) this agent can use via MCP run_plugin
+  plugins?: string[];
   // Watch: autonomous periodic monitoring
   watch?: WatchConfig;
 }
@@ -82,7 +84,7 @@ export interface AgentStep {
 // ─── Agent State (Two-Layer Model) ───────────────────────
 
 /** Smith layer: daemon lifecycle */
-export type SmithStatus = 'down' | 'active';
+export type SmithStatus = 'down' | 'starting' | 'active';
 
 /** Task layer: current work execution */
 export type TaskStatus = 'idle' | 'running' | 'done' | 'failed';

package/lib/workspace-standalone.ts

@@ -287,9 +287,18 @@ async function handleAgentsPost(id: string, body: any, res: ServerResponse): Pro
           launchInfo = resolveTerminalLaunch(agentConfig.agentId);
         } catch {}
 
-        // resolveOnly: just return launch info without side effects
+        // resolveOnly: return launch info + current session ID (no side effects)
         if (body.resolveOnly) {
-          return json(res, { ok: true, ...launchInfo });
+          let currentSessionId: string | null = null;
+          if (agentConfig.primary) {
+            try {
+              const { getFixedSession } = await import('./project-sessions.js');
+              currentSessionId = getFixedSession(orch.projectPath) || null;
+            } catch {}
+          } else {
+            currentSessionId = agentConfig.boundSessionId || null;
+          }
+          return json(res, { ok: true, ...launchInfo, currentSessionId });
         }
 
         // Primary agent: always return its fixed session, no selection
@@ -297,9 +306,12 @@ async function handleAgentsPost(id: string, body: any, res: ServerResponse): Pro
           return json(res, { ok: true, primary: true, tmuxSession: agentState.tmuxSession, fixedSession: true, ...launchInfo });
         }
 
-        if (agentState.tmuxSession) {
-          return json(res, { ok: true, alreadyOpen: true, tmuxSession: agentState.tmuxSession, ...launchInfo });
-        }
+        // No tmux has-session liveness check — it adds latency with no benefit.
+        // If the session is dead, openTerminalSession will recreate it anyway.
+
+        // Ensure tmux session exists (creates if needed, no 3s startup delay)
+        // forceRestart: kill existing tmux so launch script is rewritten with current boundSessionId
+        const tmuxSession = await orch.openTerminalSession(agentId, body.forceRestart === true);
 
         orch.setManualMode(agentId);
         // Skills call Next.js API (/api/workspace/.../smith), so use FORGE_PORT not daemon PORT
@@ -312,6 +324,7 @@ async function handleAgentsPost(id: string, body: any, res: ServerResponse): Pro
           skillsInstalled: result.installed,
           agentId,
           label: agentConfig.label,
+          tmuxSession: tmuxSession || undefined,
           ...launchInfo,
         });
       }
@@ -412,8 +425,10 @@ async function handleAgentsPost(id: string, body: any, res: ServerResponse): Pro
         return json(res, { ok: true });
       }
       case 'start_daemon': {
-        // Check active daemon count before starting
-        const activeCount = Array.from(orchestrators.values()).filter(o => o.isDaemonActive()).length;
+        // Check active daemon count before starting (include 'starting' to prevent exceeding MAX_ACTIVE)
+        const activeCount = Array.from(orchestrators.values()).filter(o =>
+          o.isDaemonActive() || Object.values(o.getAllAgentStates()).some(s => s.smithStatus === 'starting')
+        ).length;
         if (activeCount >= MAX_ACTIVE && !orch.isDaemonActive()) {
           return jsonError(res, `Maximum ${MAX_ACTIVE} active daemons. Stop agents in another workspace first.`);
         }
@@ -509,14 +524,12 @@ async function handleSmith(id: string, body: any, res: ServerResponse): Promise<
           }).trim();
         } catch {}
 
-        let gitDiffDetail = '';
-        try {
-          gitDiffDetail = execSync('git diff HEAD --name-only', {
-            cwd: orch.projectPath, encoding: 'utf-8', timeout: 5000,
-          }).trim();
-        } catch {}
-
-        const changedFiles = gitDiffDetail.split('\n').filter(Boolean);
+        // Parse file names from --stat output (lines with ' | ') instead of a second execSync
+        const changedFiles = gitDiff
+          .split('\n')
+          .filter(l => l.includes(' | '))
+          .map(l => l.split(' | ')[0].trim())
+          .filter(Boolean);
         const entry = (orch as any).agents?.get(agentId);
         const config = entry?.config;
 
@@ -584,10 +597,12 @@ async function handleSmith(id: string, body: any, res: ServerResponse): Promise<
       const target = snapshot.agents.find(a => a.label.toLowerCase() === to.toLowerCase() || a.id === to);
       if (!target) return jsonError(res, `Agent "${to}" not found. Available: ${snapshot.agents.map(a => a.label).join(', ')}`, 404);
 
-      // Resolve sender: use agentId if valid, otherwise 'user'
-      const senderId = (agentId && agentId !== 'unknown')
-        ? agentId
-        : 'user';
+      // Resolve sender: validate agentId exists in workspace, otherwise 'user'
+      const senderExists = agentId && agentId !== 'unknown' && snapshot.agents.some(a => a.id === agentId);
+      if (agentId && agentId !== 'unknown' && !senderExists) {
+        return jsonError(res, `Sender agent "${agentId}" not found in workspace`);
+      }
+      const senderId = senderExists ? agentId : 'user';
 
       // Block: if sender is currently processing a message FROM the target,
       // don't send — the result is already delivered via markMessageDone
@@ -824,7 +839,8 @@ const server = createServer(async (req, res) => {
     // Agent operations
     if (subPath === '/agents' && method === 'POST') {
       const bodyStr = await readBody(req);
-      const body = JSON.parse(bodyStr);
+      let body: any;
+      try { body = JSON.parse(bodyStr); } catch { return jsonError(res, 'Invalid JSON', 400); }
       return handleAgentsPost(id, body, res);
     }
 
@@ -840,7 +856,8 @@ const server = createServer(async (req, res) => {
     // Smith API
     if (subPath === '/smith' && method === 'POST') {
       const bodyStr = await readBody(req);
-      const body = JSON.parse(bodyStr);
+      let body: any;
+      try { body = JSON.parse(bodyStr); } catch { return jsonError(res, 'Invalid JSON', 400); }
       return handleSmith(id, body, res);
     }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aion0/forge",
-  "version": "0.5.21",
+  "version": "0.5.23",
   "description": "Unified AI workflow platform — multi-model task orchestration, persistent sessions, web terminal, remote access",
   "type": "module",
   "scripts": {

package/qa/.forge/agent-context.json

@@ -3,4 +3,4 @@
   "agentId": "qa-1774920510930",
   "agentLabel": "QA",
   "forgePort": 8403
-}
+}