@njeybe/nativ 2.3.0 → 2.3.1

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## Unreleased
+
+### Fixed
+
+- Studio Dispatch on Windows: task text with quotes or symbols like > could cut off the agent's instructions and send its output to a stray file.
+
+### Removed
+
+- The simulation endpoints (`/api/pipeline/simulation/start` and `/stop`) that fed the old Workflow Canvas.
+
 ## 2.3.0
 
 ### Changed

package/dist/runner/agent-supervisor.d.ts

@@ -209,14 +209,16 @@ export declare function buildRunnerPermissions(task: Pick<MasterPlanTask, 'verif
 /** Where a task's runner settings live: a subdirectory, because run history parses every top-level .json in runs/. */
 export declare function runnerSettingsPath(runsDir: string, taskId: string): string;
 export declare function writeRunnerSettings(runsDir: string, taskId: string, settings: RunnerPermissions): string;
+/** The instructions text handed to the agent on its standard input. */
+export declare function buildRunnerPrompt(task: MasterPlanTask): string;
 /**
- * Builds the autonomous prompt and arguments required by Claude Code CLI
- * to execute non-interactively in a headless child process.
+ * Arguments for Claude Code CLI to run headless. The prompt goes on stdin, never in this string:
+ * quotes or symbols in task text would break the shell.
  *
  * It runs in `acceptEdits` mode with the rules in `settingsFile` (see buildRunnerPermissions); it does not bypass
  * permissions. Without a settings file only the file-edit tools and read-only commands are pre-approved.
  */
-export declare function buildDefaultClaudeCommand(task: MasterPlanTask, model?: string | null, settingsFile?: string): string;
+export declare function buildDefaultClaudeCommand(_task: MasterPlanTask, model?: string | null, settingsFile?: string): string;
 /** True when a runner command asks Claude Code for its line-delimited JSON event stream. */
 export declare function usesClaudeStreamJson(command: string): boolean;
 /** Final `result` event of a Claude Code stream: grounded usage for cli runs. */

package/dist/runner/agent-supervisor.js

@@ -213,15 +213,9 @@ export function writeRunnerSettings(runsDir, taskId, settings) {
     fs.writeFileSync(file, JSON.stringify(settings, null, 2) + '\n', 'utf8');
     return file;
 }
-/**
- * Builds the autonomous prompt and arguments required by Claude Code CLI
- * to execute non-interactively in a headless child process.
- *
- * It runs in `acceptEdits` mode with the rules in `settingsFile` (see buildRunnerPermissions); it does not bypass
- * permissions. Without a settings file only the file-edit tools and read-only commands are pre-approved.
- */
-export function buildDefaultClaudeCommand(task, model, settingsFile) {
-    const parts = [
+/** The instructions text handed to the agent on its standard input. */
+export function buildRunnerPrompt(task) {
+    return [
         `Execute task ${task.id} (${task.title}).`,
         task.description ? `Description: ${task.description}.` : '',
         task.verificationCommand ? `Verify your work using: ${task.verificationCommand}.` : '',
@@ -230,11 +224,19 @@ export function buildDefaultClaudeCommand(task, model, settingsFile) {
     ]
         .filter(Boolean)
         .join(' ');
-    const escapedPrompt = parts.replace(/"/g, '\\"');
+}
+/**
+ * Arguments for Claude Code CLI to run headless. The prompt goes on stdin, never in this string:
+ * quotes or symbols in task text would break the shell.
+ *
+ * It runs in `acceptEdits` mode with the rules in `settingsFile` (see buildRunnerPermissions); it does not bypass
+ * permissions. Without a settings file only the file-edit tools and read-only commands are pre-approved.
+ */
+export function buildDefaultClaudeCommand(_task, model, settingsFile) {
     const modelFlag = model ? ` --model ${model}` : '';
     // Plain `claude -p` prints nothing until it finishes; stream-json emits every message and tool call as it happens.
     const settingsFlag = settingsFile ? ` --settings "${settingsFile}"` : '';
-    return `claude -p "${escapedPrompt}"${modelFlag} --output-format stream-json --verbose --permission-mode acceptEdits${settingsFlag}`;
+    return `claude -p${modelFlag} --output-format stream-json --verbose --permission-mode acceptEdits${settingsFlag}`;
 }
 /** True when a runner command asks Claude Code for its line-delimited JSON event stream. */
 export function usesClaudeStreamJson(command) {
@@ -1092,7 +1094,7 @@ export class AgentSupervisor extends EventEmitter {
             // stdio pipes, so we stay attached and let `taskkill /T` walk the tree.
             detached: process.platform !== 'win32',
             windowsHide: true,
-            stdio: ['ignore', 'pipe', 'pipe'],
+            stdio: ['pipe', 'pipe', 'pipe'],
             env: {
                 ...buildRunnerEnv(process.env, options.env),
                 ...(record.thinking?.budgetTokens ? { MAX_THINKING_TOKENS: String(record.thinking.budgetTokens) } : {}),
@@ -1103,6 +1105,9 @@ export class AgentSupervisor extends EventEmitter {
         });
         run.child = child;
         record.pid = child.pid ?? null;
+        // A runner that exits without reading its input must not crash us or fail the run.
+        child.stdin?.on('error', () => { });
+        child.stdin?.end(buildRunnerPrompt(task), 'utf8');
         child.stdout?.on('data', (chunk) => {
             if (run.streamJson)
                 this.consumeStreamJson(run, chunk.toString());

package/dist/server/studio-server.js

@@ -1580,128 +1580,6 @@ class PipelineEventHub {
             res.write(frame);
     }
 }
-/**
- * Interactive simulation engine for Mission Control Studio.
- * Broadcasts realistic multi-agent execution steps over the SSE stream
- * without modifying workspace files or invoking external model APIs.
- */
-class SimulationController {
-    events;
-    activeTimer = null;
-    running = false;
-    currentId = null;
-    constructor(events) {
-        this.events = events;
-    }
-    start(body = {}) {
-        this.stop();
-        const id = `sim-${Date.now()}`;
-        this.currentId = id;
-        this.running = true;
-        const speed = Math.max(1, Math.min(5, Number(body.speedMultiplier) || 1));
-        const scenario = String(body.scenario || 'default');
-        const steps = [
-            // Step 1: Foundation task (qa-tester)
-            { taskId: 'task-01', phase: 'spawning_worktree', log: '[qa-tester] Provisioning isolated worktree branch for task-01...\n', delay: 400 },
-            { taskId: 'task-01', phase: 'running', log: '[qa-tester] Validating dependencies, project manifest and baseline checks...\n', delay: 600 },
-            { taskId: 'task-01', phase: 'verifying', log: '[gatekeeper] Running verification command: npm test\n✔ Baseline contracts validated\n', delay: 600 },
-            { taskId: 'task-01', phase: 'completed', log: '[qa-tester] Task task-01 passed verification and merged.\n', delay: 400 },
-            // Step 2: Parallel execution (backend & frontend)
-            { taskId: 'task-studio-sse-api', phase: 'spawning_worktree', log: '[backend] Initializing git worktree .nativ/worktrees/task-studio-sse-api...\n', delay: 300 },
-            { taskId: 'task-studio-light-ui', phase: 'spawning_worktree', log: '[frontend] Initializing git worktree .nativ/worktrees/task-studio-light-ui...\n', delay: 300 },
-            { taskId: 'task-studio-sse-api', phase: 'running', log: '[backend] Implementing real-time event pipeline fanout & simulation endpoints in studio-server.ts...\n', delay: 700 },
-            { taskId: 'task-studio-light-ui', phase: 'running', log: '[frontend] Constructing SVG Workflow Canvas with Bezier cables and floating playback bar...\n', delay: 700 },
-            { taskId: 'task-studio-sse-api', phase: 'verifying', log: '[gatekeeper] Executing test-studio-pipeline-api.mjs...\n✔ All API endpoints passed\n', delay: 600 },
-            { taskId: 'task-studio-sse-api', phase: 'merging', log: '[backend] Merging worktree branch to main...\n', delay: 300 },
-            { taskId: 'task-studio-sse-api', phase: 'completed', log: '[backend] SSE pipeline stream online.\n', delay: 400 },
-            { taskId: 'task-studio-light-ui', phase: 'verifying', log: '[gatekeeper] Executing test-studio-light-ui.mjs...\n✔ Zero-emoji compliance and Canvas DOM validated\n', delay: 600 },
-            { taskId: 'task-studio-light-ui', phase: 'merging', log: '[frontend] Merging worktree branch to main...\n', delay: 300 },
-            { taskId: 'task-studio-light-ui', phase: 'completed', log: '[frontend] Workflow Canvas ready.\n', delay: 400 },
-            // Step 3: End-to-end integration & verification
-            { taskId: 'task-studio-e2e-verify', phase: 'spawning_worktree', log: '[qa-tester] Spawning E2E integration test runner...\n', delay: 300 },
-            { taskId: 'task-studio-e2e-verify', phase: 'running', log: '[qa-tester] Running comprehensive integration test matrix...\n', delay: 800 },
-            ...(scenario === 'circuit_breaker_heal'
-                ? [
-                    { taskId: 'task-studio-e2e-verify', phase: 'failed', stream: 'stderr', log: '[qa-tester] Invariant violation detected: test mock failure. Circuit breaker tripped.\n', delay: 600 },
-                ]
-                : [
-                    { taskId: 'task-studio-e2e-verify', phase: 'verifying', log: '[gatekeeper] Full E2E suite passed.\n', delay: 600 },
-                    { taskId: 'task-studio-e2e-verify', phase: 'completed', log: '[qa-tester] All milestones verified.\n', delay: 400 },
-                ]),
-        ];
-        let stepIndex = 0;
-        const runNext = () => {
-            if (!this.running || stepIndex >= steps.length) {
-                this.running = false;
-                return;
-            }
-            const step = steps[stepIndex++];
-            const record = {
-                runId: `${id}-${step.taskId}`,
-                taskId: step.taskId,
-                status: step.phase,
-                engine: 'native',
-                model: 'claude-opus-5-5 (simulation)',
-                thinking: { budget: 2048, effort: 'medium', budgetTokens: 2048 },
-                usage: {
-                    model: 'claude-opus-5-5 (simulation)',
-                    turns: 3,
-                    inputTokens: 1400,
-                    outputTokens: 380,
-                    cacheCreationTokens: 500,
-                    cacheReadTokens: 4000,
-                    thinkingTokens: 120,
-                    costUsd: 0.011,
-                },
-                pid: 99000 + stepIndex,
-                branch: `nativ/${step.taskId}`,
-                worktreeDir: `.nativ/worktrees/${step.taskId}`,
-                command: 'simulation',
-                startedAt: new Date(Date.now() - 2000).toISOString(),
-                endedAt: step.phase === 'completed' || step.phase === 'failed' ? new Date().toISOString() : null,
-                durationMs: 2500,
-                exitCode: step.phase === 'completed' ? 0 : step.phase === 'failed' ? 1 : null,
-                signal: null,
-                timeoutSeconds: 300,
-                timedOut: false,
-                logBytes: (step.log || '').length,
-                logFile: `.nativ/logs/${step.taskId}.log`,
-                error: step.phase === 'failed' ? (step.log || 'Task failed') : null,
-                abortReason: null,
-                verification: step.phase === 'completed' ? { command: 'npm test', success: true, exitCode: 0, durationMs: 400, skipped: false } : null,
-            };
-            this.events.publish('runner_status', runView(record));
-            if (step.log) {
-                this.events.publish('runner_log', {
-                    runId: `${id}-${step.taskId}`,
-                    taskId: step.taskId,
-                    stream: step.stream ?? 'stdout',
-                    chunk: step.log,
-                    at: new Date().toISOString(),
-                });
-            }
-            this.activeTimer = setTimeout(runNext, Math.max(80, Math.round(step.delay / speed)));
-        };
-        this.activeTimer = setTimeout(runNext, Math.round(150 / speed));
-        return {
-            ok: true,
-            simulationId: id,
-            scenario,
-            speed,
-            status: 'running',
-        };
-    }
-    stop() {
-        if (this.activeTimer) {
-            clearTimeout(this.activeTimer);
-            this.activeTimer = null;
-        }
-        const wasRunning = this.running;
-        this.running = false;
-        this.currentId = null;
-        return { ok: true, status: 'stopped', wasRunning };
-    }
-}
 // ─── Server ────────────────────────────────────────────────────────────────────
 export function createStudioServer(options = {}) {
     const cwd = options.cwd ?? process.cwd();
@@ -1742,7 +1620,6 @@ export function createStudioServer(options = {}) {
     supervisor.on('runner_status', (record) => events.publish('runner_status', runView(record)));
     supervisor.on('runner_log', (entry) => events.publish('runner_log', entry));
     supervisor.on('runner_token_usage', (usage) => events.publish('runner_token_usage', usage));
-    const simulation = new SimulationController(events);
     // Tier 1 strategist: settings and counters live for this server session only.
     const liaison = new Tier1Liaison(root, options.triage);
     const triageInFlight = new Map();
@@ -1842,10 +1719,6 @@ export function createStudioServer(options = {}) {
                     return sendJson(res, 200, handleEscalations(root, url.searchParams));
                 case 'POST /api/pipeline/escalations/resolve':
                     return sendJson(res, 200, await handleEscalationResolve(root, await readJsonBody(req)));
-                case 'POST /api/pipeline/simulation/start':
-                    return sendJson(res, 200, simulation.start(await readJsonBody(req)));
-                case 'POST /api/pipeline/simulation/stop':
-                    return sendJson(res, 200, simulation.stop());
                 case 'POST /api/pipeline/triage/evaluate':
                     return sendJson(res, 200, await handleTriageEvaluate(root, liaison, triageInFlight, await readJsonBody(req)));
                 case 'GET /api/pipeline/triage/status':
@@ -1863,8 +1736,7 @@ export function createStudioServer(options = {}) {
                     '/api/pipeline/tasks/abort', '/api/pipeline/tasks/runs', '/api/pipeline/tasks/logs', '/api/pipeline/worktrees',
                     '/api/pipeline/worktrees/action', '/api/pipeline/worktrees/diff', '/api/pipeline/benchmarks', '/api/pipeline/benchmarks/run',
                     '/api/pipeline/telemetry/detailed', '/api/pipeline/escalations', '/api/pipeline/escalations/resolve',
-                    '/api/pipeline/simulation/start', '/api/pipeline/simulation/stop', '/api/pipeline/triage/evaluate',
-                    '/api/pipeline/triage/status', '/api/pipeline/triage/config',
+                    '/api/pipeline/triage/evaluate', '/api/pipeline/triage/status', '/api/pipeline/triage/config',
                 ];
                 if (known.includes(url.pathname.replace(/\/+$/, '')))
                     throw new HttpError(405, 'METHOD_NOT_ALLOWED', `${req.method} not allowed on ${url.pathname}`);
@@ -1884,7 +1756,6 @@ export function createStudioServer(options = {}) {
     // Open SSE streams and in-flight agent runners would otherwise keep close() waiting forever.
     const closeServer = server.close.bind(server);
     server.close = ((callback) => {
-        simulation.stop();
         supervisor.shutdown('Studio server shutting down');
         events.close();
         return closeServer(callback);

package/dist/server/studio-ui/shell/dispatch.js

@@ -59,15 +59,9 @@ export const dispatchScript = String.raw `  // ─── Autonomous dispatch: PO
 
   /** Mirrors buildDefaultClaudeCommand() in agent-supervisor.ts so the preview matches what the server runs. */
   function defaultClaudeCommand(t) {
-    var prompt = [
-      'Execute task ' + t.id + ' (' + t.title + ').',
-      t.description ? 'Description: ' + t.description + '.' : '',
-      t.verificationCommand ? 'Verify your work using: ' + t.verificationCommand + '.' : '',
-      'Start by running: nativ task start ' + t.id + '. When finished and verified, run: nativ task complete ' + t.id + '.',
-      'If blocked, follow the Human-Centric Communication Protocol in CLAUDE.md: explain the user experience symptom, root cause in plain English, and clear options without technical jargon.'
-    ].filter(Boolean).join(' ');
-    // No permission bypass: acceptEdits plus the per-task allowlist the server writes to .nativ/runs/permissions/<task>.json.
-    return 'claude -p "' + prompt.replace(/"/g, '\\"') + '" --output-format stream-json --verbose --permission-mode acceptEdits --settings ".nativ/runs/permissions/' + t.id + '.json"';
+    // The task prompt goes to the agent on stdin, so it is not part of the command.
+    return 'claude -p --output-format stream-json --verbose ' +
+      '--permission-mode acceptEdits --settings ".nativ/runs/permissions/' + t.id + '.json"';
   }
 
   function syncDispatchSwitches() {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@njeybe/nativ",
-  "version": "2.3.0",
+  "version": "2.3.1",
   "description": "Role-based multi-agent workflow harness: Claude Code by default, other AI agents pluggable",
   "type": "module",
   "main": "dist/index.js",

package/plugin/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "nativ",
   "displayName": "nativ",
-  "version": "2.3.0",
+  "version": "2.3.1",
   "description": "Role-based multi-agent workflow for Claude Code: architect, worker and verifier agents, task-scope enforcement, and the nativ MCP server.",
   "author": {
     "name": "JB Natividad"