chore: bump version to 0.1431

This commit is contained in:
arkon
2026-01-29 13:22:02 +01:00
parent 8ed1c218f4
commit 7b866c1998
14 changed files with 29 additions and 3512 deletions
+1 -21
View File
@@ -16,7 +16,7 @@ When user says "COM":
1. Increment version in BOTH `package.json` AND `CLAUDE.md`
2. Run: `git add -A && git commit -m "chore: bump version to X.XXXX" && git push && npm run build && systemctl --user restart claudeman-web`
**Version**: 0.1430 (must match `package.json`)
**Version**: 0.1431 (must match `package.json`)
## Project Overview
@@ -56,7 +56,6 @@ journalctl --user -u claudeman-web -f
| Binary | Purpose |
|--------|---------|
| `claudeman` | Main CLI and web server |
| `claudeman-mcp` | MCP server for Claude Desktop integration |
## Architecture
@@ -191,7 +190,6 @@ Use `LRUMap` for bounded caches with eviction, `StaleExpirationMap` for TTL-base
| Topic | Location |
|-------|----------|
| **Respawn state machine** | `docs/respawn-state-machine.md` |
| **Spawn agent protocol** | `docs/spawn-protocol.md` |
| **Ralph Loop guide** | `docs/ralph-wiggum-guide.md` |
| **Claude Code hooks** | `docs/claude-code-hooks-reference.md` |
| **Browser/E2E testing** | `docs/browser-testing-guide.md` |
@@ -208,24 +206,6 @@ Use `LRUMap` for bounded caches with eviction, `StaleExpirationMap` for TTL-base
| **Execution system** | `src/config/execution-limits.ts` |
| **Agent prompts** | `src/prompts/` directory |
## MCP Server Integration
The `claudeman-mcp` binary provides Model Context Protocol integration for Claude Desktop:
```json
// Add to ~/.claude/claude_desktop_config.json
{
"mcpServers": {
"claudeman-spawn": {
"command": "node",
"args": ["/path/to/claudeman/dist/mcp-server.js"]
}
}
}
```
This enables Claude Desktop to spawn and manage agents via MCP tools.
## Scripts
| Script | Purpose |
-55
View File
@@ -126,43 +126,6 @@ Claudeman detects and tracks Ralph Loops and Todos inside Claude Code:
---
### 🤖 Spawn1337: Autonomous Agent Protocol
Spawn full-power Claude agents that run independently in their own screen sessions:
```
Parent Session → <spawn1337>task.md</spawn1337>
→ SpawnDetector parses tag
→ Orchestrator creates agent directory
→ Spawns Claude in its own screen session
→ Agent works autonomously
→ Reports result via <promise>PHRASE</promise>
→ Parent notified via SSE
```
**Features:**
- **Resource governance**: Budget limits (tokens + cost), timeout enforcement, graceful shutdown
- **Agent trees**: Agents can spawn children (max depth: 3)
- **Communication**: Filesystem-based message bus between parent and child
- **Max 5 concurrent** agents with queuing for overflow
```yaml
# Task spec format (YAML frontmatter in .md file)
---
agentId: my-agent-001
name: My Agent
type: implement
priority: high
maxTokens: 150000
maxCost: 0.50
timeoutMinutes: 15
completionPhrase: AGENT_DONE
---
Implement the feature described below...
```
---
### 👁️ Live Agent Visualization
**Watch your agents work in real-time.** Claudeman monitors Claude Code's background agents (the `Task` tool) and displays them in draggable floating windows with Matrix-style connection lines.
@@ -342,16 +305,6 @@ claudeman web
| `GET` | `/api/sessions/:id/ralph-state` | Get loop state + todos |
| `POST` | `/api/sessions/:id/ralph-config` | Configure tracking |
### Spawn Agents
| Method | Endpoint | Description |
|--------|----------|-------------|
| `GET` | `/api/spawn/agents` | List all agents |
| `GET` | `/api/spawn/agents/:id` | Agent status + progress |
| `GET` | `/api/spawn/agents/:id/result` | Agent result |
| `POST` | `/api/spawn/agents/:id/message` | Send message to agent |
| `POST` | `/api/spawn/agents/:id/cancel` | Cancel agent |
| `POST` | `/api/spawn/trigger` | Programmatic spawn |
### Subagents (Claude Code Background Agents)
| Method | Endpoint | Description |
|--------|----------|-------------|
@@ -395,12 +348,10 @@ flowchart TB
S1["Session (PTY)"]
S2["Session (PTY)"]
RC["Respawn Controller"]
SO["Spawn Orchestrator"]
end
subgraph Detection["Detection Layer"]
RT["Ralph Tracker"]
SD["Spawn Detector"]
SW["Subagent Watcher<br/><small>~/.claude/projects/*/subagents</small>"]
end
@@ -411,8 +362,6 @@ flowchart TB
subgraph External["External"]
CLI["Claude CLI"]
A1["Agent 1<br/><small>(screen)</small>"]
A2["Agent 2<br/><small>(screen)</small>"]
BG["Background Agents<br/><small>(Task tool)</small>"]
end
end
@@ -425,10 +374,6 @@ flowchart TB
SM --> RC
SM --> SS
S1 --> RT
S1 --> SD
SD --> SO
SO --> A1
SO --> A2
S1 --> SCR
S2 --> SCR
RC --> SCR
BIN
View File
Binary file not shown.

After

Width:  |  Height:  |  Size: 914 KiB

-80
View File
@@ -1,80 +0,0 @@
# Spawn1337 Protocol (Autonomous Agents)
Spawned agents are full-power Claude sessions running in their own screen sessions. They communicate via a filesystem-based message bus and signal completion via RalphTracker's `<promise>` mechanism.
## Primary Interface: MCP Server
The `claudeman-mcp` binary exposes spawn tools to Claude Code via MCP protocol, replacing the legacy terminal-tag-parsing approach (SpawnDetector).
### MCP Tools
| Tool | Description |
|------|-------------|
| `spawn_agent` | Spawn a new autonomous agent (builds task spec from parameters) |
| `list_agents` | List all agents (active + completed + queued) |
| `get_agent_status` | Get detailed agent status + progress |
| `get_agent_result` | Read a completed agent's result |
| `send_agent_message` | Send a message to a running agent |
| `cancel_agent` | Cancel a running agent |
### Environment Variables
| Variable | Description |
|----------|-------------|
| `CLAUDEMAN_API_URL` | Base URL for the Claudeman API (default: `http://localhost:3000`) |
| `CLAUDEMAN_SESSION_ID` | Session ID of the calling Claude session |
## Protocol Flow
```
Claude calls spawn_agent MCP tool
→ MCP server builds task spec YAML
→ POST /api/spawn/trigger with task spec
→ SpawnOrchestrator creates agent directory: ~/claudeman-cases/spawn-<agentId>/
→ Spawns interactive Claude session in screen
→ Injects initial prompt via writeViaScreen()
→ Agent works autonomously, writes progress to spawn-comms/
→ RalphTracker detects <promise>PHRASE</promise> on child
→ Orchestrator reads result.md, notifies parent via SSE
```
## Agent Directory Structure
Each agent gets: `~/claudeman-cases/spawn-<agentId>/`
```
spawn-<agentId>/
├── CLAUDE.md # Generated from spawn-claude-md.ts
├── spawn-comms/
│ ├── task.md # Task specification
│ ├── progress.json # Current progress state
│ ├── result.md # Final result (on completion)
│ └── messages/ # Inter-agent messaging
└── workspace/ # Symlinked context files
```
## Resource Governance
| Limit | Value |
|-------|-------|
| Max concurrent agents | 5 |
| Max depth | 3 |
| Default timeout | 30min |
| Max timeout | 120min |
| Budget warning | 80% |
| Graceful shutdown | 100% |
| Force kill | 110% |
## Agent Tree
Agents can spawn children. Sessions track `parentAgentId` and `childAgentIds`. Cancelling a parent cascades to all children.
## Key Source Files
| File | Purpose |
|------|---------|
| `src/mcp-server.ts` | MCP server binary exposing spawn tools |
| `src/spawn-orchestrator.ts` | Full agent lifecycle: spawn, monitor, budget, queue, cleanup |
| `src/spawn-claude-md.ts` | Generates CLAUDE.md for spawned agent sessions |
| `src/spawn-types.ts` | Types, YAML parser, factory functions |
| `src/spawn-detector.ts` | Legacy: detects `<spawn1337>` tags in terminal output |
+3 -5
View File
@@ -1,16 +1,15 @@
{
"name": "claudeman",
"version": "0.1430",
"version": "0.1431",
"description": "The missing control plane for Claude Code - run 20 autonomous agents with real-time monitoring and session persistence",
"type": "module",
"main": "dist/index.js",
"bin": {
"claudeman": "./dist/index.js",
"claudeman-mcp": "./dist/mcp-server.js"
"claudeman": "./dist/index.js"
},
"scripts": {
"postinstall": "node scripts/postinstall.js",
"build": "tsc && chmod +x dist/index.js dist/mcp-server.js && mkdir -p dist/web dist/templates && cp -r src/web/public dist/web/ && cp src/templates/case-template.md dist/templates/",
"build": "tsc && chmod +x dist/index.js && mkdir -p dist/web dist/templates && cp -r src/web/public dist/web/ && cp src/templates/case-template.md dist/templates/",
"start": "node dist/index.js",
"dev": "tsx src/index.ts",
"web": "node dist/index.js web",
@@ -39,7 +38,6 @@
"license": "MIT",
"dependencies": {
"@fastify/static": "^8.0.0",
"@modelcontextprotocol/sdk": "^1.25.3",
"chalk": "^5.3.0",
"chokidar": "^3.6.0",
"commander": "^12.1.0",
-370
View File
@@ -1,370 +0,0 @@
#!/usr/bin/env node
/**
* @fileoverview MCP Server for Claudeman Spawn1337 protocol.
*
* Exposes spawn capabilities as native MCP tools that Claude Code can call
* directly, replacing the terminal-tag-parsing approach (SpawnDetector).
*
* Tools:
* - spawn_agent: Spawn a new autonomous agent
* - list_agents: List all agents (active + completed)
* - get_agent_status: Get detailed agent status
* - get_agent_result: Read a completed agent's result
* - send_agent_message: Send a message to a running agent
* - cancel_agent: Cancel a running agent
*
* Environment:
* - CLAUDEMAN_API_URL: Base URL for the Claudeman API (default: http://localhost:3000)
* - CLAUDEMAN_SESSION_ID: Session ID of the calling Claude session
*
* @module mcp-server
*/
import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
import { z } from 'zod';
// ========== Configuration ==========
const API_URL = process.env.CLAUDEMAN_API_URL || 'http://localhost:3000';
const SESSION_ID = process.env.CLAUDEMAN_SESSION_ID || '';
// ========== API Helper ==========
/**
* Make an HTTP request to the Claudeman API.
*/
async function apiRequest(method: string, path: string, body?: unknown): Promise<{ status: number; data: unknown }> {
const url = `${API_URL}${path}`;
const options: RequestInit = {
method,
headers: { 'Content-Type': 'application/json' },
};
if (body !== undefined) {
options.body = JSON.stringify(body);
}
const response = await fetch(url, options);
const data = await response.json();
return { status: response.status, data };
}
// ========== YAML Construction ==========
/**
* Build a YAML frontmatter + body task spec from structured parameters.
*/
export function buildTaskSpec(params: {
agentId: string;
name: string;
instructions: string;
type?: string;
priority?: string;
maxTokens?: number;
maxCost?: number;
timeoutMinutes?: number;
canModifyParentFiles?: boolean;
contextFiles?: string[];
dependsOn?: string[];
completionPhrase?: string;
outputFormat?: string;
successCriteria?: string;
workingDir?: string;
}): string {
const lines: string[] = ['---'];
lines.push(`agentId: ${params.agentId}`);
lines.push(`name: ${params.name}`);
if (params.type) lines.push(`type: ${params.type}`);
if (params.priority) lines.push(`priority: ${params.priority}`);
if (params.maxTokens != null) lines.push(`maxTokens: ${params.maxTokens}`);
if (params.maxCost != null) lines.push(`maxCost: ${params.maxCost}`);
if (params.timeoutMinutes != null) lines.push(`timeoutMinutes: ${params.timeoutMinutes}`);
if (params.canModifyParentFiles != null) lines.push(`canModifyParentFiles: ${params.canModifyParentFiles}`);
if (params.completionPhrase) lines.push(`completionPhrase: ${params.completionPhrase}`);
if (params.outputFormat) lines.push(`outputFormat: ${params.outputFormat}`);
if (params.successCriteria) lines.push(`successCriteria: "${params.successCriteria.replace(/"/g, '\\"')}"`);
if (params.workingDir) lines.push(`workingDir: ${params.workingDir}`);
if (params.contextFiles && params.contextFiles.length > 0) {
lines.push(`contextFiles: [${params.contextFiles.join(', ')}]`);
}
if (params.dependsOn && params.dependsOn.length > 0) {
lines.push(`dependsOn: [${params.dependsOn.join(', ')}]`);
}
lines.push('---');
lines.push('');
lines.push(params.instructions);
return lines.join('\n');
}
// ========== MCP Server Setup ==========
const server = new McpServer({
name: 'claudeman-spawn',
version: '1.0.0',
});
// ---- spawn_agent ----
server.tool(
'spawn_agent',
'Spawn a new autonomous Claude agent to handle a subtask. The agent runs in its own session with full capabilities.',
{
agentId: z.string().describe('Unique identifier for the agent (e.g., "research-auth-001")'),
name: z.string().describe('Human-readable name for the agent'),
instructions: z.string().describe('Detailed task instructions for the agent (markdown)'),
type: z.enum(['explore', 'implement', 'test', 'review', 'refactor', 'research', 'generate', 'fix', 'general']).optional().describe('Task type/category'),
priority: z.enum(['low', 'normal', 'high', 'critical']).optional().describe('Priority for queue ordering'),
maxTokens: z.number().optional().describe('Maximum token budget (input + output combined)'),
maxCost: z.number().optional().describe('Maximum cost in USD'),
timeoutMinutes: z.number().optional().describe('Maximum runtime in minutes (max: 120)'),
canModifyParentFiles: z.boolean().optional().describe('Whether agent can modify files in parent project directory'),
contextFiles: z.array(z.string()).optional().describe('Files to symlink into agent workspace as context'),
dependsOn: z.array(z.string()).optional().describe('Agent IDs that must complete before this one starts'),
completionPhrase: z.string().optional().describe('Phrase agent outputs when finished (default: auto-generated)'),
outputFormat: z.enum(['markdown', 'json', 'code', 'structured', 'freeform']).optional().describe('Expected output format'),
successCriteria: z.string().optional().describe('Success criteria included in agent instructions'),
workingDir: z.string().optional().describe('Working directory (relative to parent, or absolute)'),
},
async (params) => {
if (!SESSION_ID) {
return {
content: [{ type: 'text', text: 'Error: CLAUDEMAN_SESSION_ID not set. This tool must be run within a Claudeman-managed session.' }],
isError: true,
};
}
const taskSpec = buildTaskSpec(params);
try {
const { status, data } = await apiRequest('POST', '/api/spawn/trigger', {
parentSessionId: SESSION_ID,
taskContent: taskSpec,
});
if (status >= 400) {
const errorData = data as { error?: { code?: string; details?: string } };
const errorMsg = errorData.error?.details || errorData.error?.code || 'Unknown error';
return {
content: [{ type: 'text', text: `Error spawning agent: ${errorMsg}` }],
isError: true,
};
}
const result = data as { success: boolean; data?: { agentId: string } };
const agentId = result.data?.agentId || params.agentId;
return {
content: [{ type: 'text', text: `Agent spawned successfully.\n\nAgent ID: ${agentId}` }],
};
} catch (err) {
return {
content: [{ type: 'text', text: `Error: Could not connect to Claudeman API at ${API_URL}. Is the web server running?\n\n${err instanceof Error ? err.message : String(err)}` }],
isError: true,
};
}
}
);
// ---- list_agents ----
server.tool(
'list_agents',
'List all spawn agents (active, queued, and completed).',
{},
async () => {
try {
const { status, data } = await apiRequest('GET', '/api/spawn/agents');
if (status >= 400) {
return {
content: [{ type: 'text', text: `Error listing agents: ${JSON.stringify(data)}` }],
isError: true,
};
}
const agents = data as Array<{ agentId: string; name: string; status: string; type: string; priority: string }>;
if (agents.length === 0) {
return { content: [{ type: 'text', text: 'No agents found.' }] };
}
const summary = agents.map(a => `- ${a.agentId} (${a.name}): ${a.status} [${a.type}, ${a.priority}]`).join('\n');
return { content: [{ type: 'text', text: `Agents (${agents.length}):\n\n${summary}` }] };
} catch (err) {
return {
content: [{ type: 'text', text: `Error: Could not connect to Claudeman API at ${API_URL}.\n\n${err instanceof Error ? err.message : String(err)}` }],
isError: true,
};
}
}
);
// ---- get_agent_status ----
server.tool(
'get_agent_status',
'Get detailed status and progress of a specific agent.',
{
agentId: z.string().describe('The agent ID to query'),
},
async ({ agentId }) => {
try {
const { status, data } = await apiRequest('GET', `/api/spawn/agents/${encodeURIComponent(agentId)}`);
if (status === 404) {
return {
content: [{ type: 'text', text: `Agent not found: ${agentId}` }],
isError: true,
};
}
if (status >= 400) {
return {
content: [{ type: 'text', text: `Error getting agent status: ${JSON.stringify(data)}` }],
isError: true,
};
}
return { content: [{ type: 'text', text: JSON.stringify(data, null, 2) }] };
} catch (err) {
return {
content: [{ type: 'text', text: `Error: Could not connect to Claudeman API at ${API_URL}.\n\n${err instanceof Error ? err.message : String(err)}` }],
isError: true,
};
}
}
);
// ---- get_agent_result ----
server.tool(
'get_agent_result',
'Read the result of a completed agent. Returns the agent\'s output and metadata.',
{
agentId: z.string().describe('The agent ID whose result to read'),
},
async ({ agentId }) => {
try {
const { status, data } = await apiRequest('GET', `/api/spawn/agents/${encodeURIComponent(agentId)}/result`);
if (status === 404) {
return {
content: [{ type: 'text', text: `Agent or result not found: ${agentId}. The agent may not have completed yet.` }],
isError: true,
};
}
if (status >= 400) {
return {
content: [{ type: 'text', text: `Error getting agent result: ${JSON.stringify(data)}` }],
isError: true,
};
}
// Result could be a string (raw markdown) or an object
const resultText = typeof data === 'string' ? data : JSON.stringify(data, null, 2);
return { content: [{ type: 'text', text: resultText }] };
} catch (err) {
return {
content: [{ type: 'text', text: `Error: Could not connect to Claudeman API at ${API_URL}.\n\n${err instanceof Error ? err.message : String(err)}` }],
isError: true,
};
}
}
);
// ---- send_agent_message ----
server.tool(
'send_agent_message',
'Send a message to a running agent. The message is written to the agent\'s communication channel.',
{
agentId: z.string().describe('The agent ID to message'),
message: z.string().describe('The message content (markdown)'),
},
async ({ agentId, message }) => {
try {
const { status, data } = await apiRequest('POST', `/api/spawn/agents/${encodeURIComponent(agentId)}/message`, {
content: message,
sender: 'parent',
});
if (status === 404) {
return {
content: [{ type: 'text', text: `Agent not found: ${agentId}` }],
isError: true,
};
}
if (status >= 400) {
return {
content: [{ type: 'text', text: `Error sending message: ${JSON.stringify(data)}` }],
isError: true,
};
}
return { content: [{ type: 'text', text: `Message sent to agent ${agentId}.` }] };
} catch (err) {
return {
content: [{ type: 'text', text: `Error: Could not connect to Claudeman API at ${API_URL}.\n\n${err instanceof Error ? err.message : String(err)}` }],
isError: true,
};
}
}
);
// ---- cancel_agent ----
server.tool(
'cancel_agent',
'Cancel a running agent. Sends a graceful shutdown signal.',
{
agentId: z.string().describe('The agent ID to cancel'),
reason: z.string().optional().describe('Reason for cancellation'),
},
async ({ agentId, reason }) => {
try {
const { status, data } = await apiRequest('POST', `/api/spawn/agents/${encodeURIComponent(agentId)}/cancel`, {
reason: reason || 'Cancelled by parent session',
});
if (status === 404) {
return {
content: [{ type: 'text', text: `Agent not found: ${agentId}` }],
isError: true,
};
}
if (status >= 400) {
return {
content: [{ type: 'text', text: `Error cancelling agent: ${JSON.stringify(data)}` }],
isError: true,
};
}
return { content: [{ type: 'text', text: `Agent ${agentId} cancel request sent.` }] };
} catch (err) {
return {
content: [{ type: 'text', text: `Error: Could not connect to Claudeman API at ${API_URL}.\n\n${err instanceof Error ? err.message : String(err)}` }],
isError: true,
};
}
}
);
// ========== Start Server ==========
async function main(): Promise<void> {
const transport = new StdioServerTransport();
await server.connect(transport);
}
main().catch((err) => {
console.error('MCP server failed to start:', err);
process.exit(1);
});
-154
View File
@@ -1,154 +0,0 @@
/**
* @fileoverview Agent CLAUDE.md Generator for spawn1337 protocol.
*
* Generates a comprehensive CLAUDE.md for each spawned agent that tells it:
* - What its task is
* - How to communicate progress
* - How to signal completion
* - What constraints it has
* - How to read/write messages
*
* @module spawn-claude-md
*/
import type { SpawnTask } from './spawn-types.js';
/**
* Generate a CLAUDE.md file for a spawned agent.
*
* This CLAUDE.md gives the agent full context about:
* - Its identity and task
* - Communication protocol (progress, messages, result)
* - Resource constraints (timeout, tokens, cost)
* - Working directory and available context files
*
* @param task - The full parsed task specification
* @param commsDir - Absolute path to the communication directory
* @param agentWorkingDir - Absolute path to the agent's working directory
* @returns The generated CLAUDE.md content
*/
export function generateAgentClaudeMd(task: SpawnTask, commsDir: string, agentWorkingDir: string): string {
const spec = task.spec;
const constraintLines: string[] = [];
constraintLines.push(`- Timeout: ${spec.timeoutMinutes} minutes`);
if (spec.maxTokens) constraintLines.push(`- Token budget: ${spec.maxTokens.toLocaleString()} tokens`);
if (spec.maxCost) constraintLines.push(`- Cost budget: $${spec.maxCost.toFixed(2)}`);
if (!spec.canModifyParentFiles) {
constraintLines.push('- DO NOT modify files outside your workspace');
} else {
constraintLines.push('- You MAY modify files in the parent project directory');
}
constraintLines.push(`- Output format: ${spec.outputFormat}`);
const contextSection = spec.contextFiles && spec.contextFiles.length > 0
? `\nContext files available in workspace:\n${spec.contextFiles.map(f => `- ${f}`).join('\n')}`
: '';
const progressSection = spec.progressIntervalSeconds > 0
? `### Progress Reporting
Update \`${commsDir}/progress.json\` every ~${spec.progressIntervalSeconds} seconds with your current status:
\`\`\`json
{
"phase": "current phase description",
"percentComplete": 45,
"currentAction": "What you are doing right now",
"subtasks": [
{"description": "Subtask 1", "status": "completed"},
{"description": "Subtask 2", "status": "in_progress"}
],
"filesModified": ["file1.ts", "file2.ts"],
"tokensUsed": 0,
"costSoFar": 0,
"updatedAt": ${Date.now()}
}
\`\`\``
: '### Progress Reporting\n\nProgress reporting is disabled for this task.';
return `# Agent: ${spec.name}
## Your Identity
You are an autonomous agent (ID: \`${spec.agentId}\`) spawned by a parent Claude session.
You are running in your own screen session with full Claude Code capabilities.
Type: ${spec.type} | Priority: ${spec.priority} | Depth: ${task.depth}
## Task
${task.instructions}
## Success Criteria
${spec.successCriteria || 'Complete the task as described above.'}
## Communication Protocol
${progressSection}
### Check for Messages
Periodically check \`${commsDir}/messages/\` for instructions from the parent.
Files are named \`NNN-parent.md\` (from parent) or \`NNN-agent.md\` (from you).
Read any new \`*-parent.md\` files for additional instructions or clarifications.
To send a message back to the parent, create a file like:
\`${commsDir}/messages/002-agent.md\`
### Write Result
When complete, write your final result to \`${commsDir}/result.md\` with YAML frontmatter:
\`\`\`markdown
---
status: completed
summary: "Brief 1-3 sentence summary of what you accomplished"
filesChanged:
- path: relative/path/to/file.ts
action: modified
summary: "What was changed"
---
## Detailed Output
Your full output, analysis, or report here.
\`\`\`
Valid status values: \`completed\`, \`failed\`
### Signal Completion
After writing result.md, output this EXACT phrase to signal you are done:
<promise>${spec.completionPhrase}</promise>
**IMPORTANT**: Only output the completion phrase AFTER you have written result.md.
The completion phrase triggers the orchestrator to read your result and clean up.
## Constraints
${constraintLines.join('\n')}
## Working Directory
Your workspace is: \`${agentWorkingDir}\`
${contextSection}
## Important Notes
- Work autonomously - do not ask for user input
- Focus exclusively on the task described above
- If you encounter errors, document them in result.md with status: failed
- Do not modify this CLAUDE.md file
- Stay within your resource constraints
`;
}
/**
* Build the initial prompt injected into the agent session via writeViaScreen().
* Intentionally brief - all detail is in the CLAUDE.md.
*/
export function buildInitialPrompt(task: SpawnTask): string {
return `Read your CLAUDE.md file for complete task instructions, communication protocol, and constraints. Begin working on the task immediately. Report progress to spawn-comms/progress.json periodically. When complete, write your result to spawn-comms/result.md and then output your completion phrase: <promise>${task.spec.completionPhrase}</promise>`;
}
-964
View File
@@ -1,964 +0,0 @@
/**
* @fileoverview Spawn Orchestrator - Full lifecycle management for spawned agents.
*
* Manages:
* - Agent creation from task spec files
* - Directory setup (CLAUDE.md, comms, workspace)
* - Session spawning via screen
* - Progress monitoring and timeout enforcement
* - Resource governance (tokens, cost, depth)
* - Bidirectional communication
* - Result collection and cleanup
* - Queue management with priority ordering
*
* @module spawn-orchestrator
*/
import { EventEmitter } from 'node:events';
import { join, resolve, isAbsolute } from 'node:path';
import { existsSync, mkdirSync, writeFileSync, readFileSync, readdirSync, statSync, symlinkSync } from 'node:fs';
import { v4 as uuidv4 } from 'uuid';
import {
type SpawnOrchestratorConfig,
type SpawnTask,
type AgentContext,
type AgentProgress,
type AgentStatusReport,
type SpawnResult,
type SpawnTrackerState,
type SpawnMessage,
type SpawnPersistedState,
createDefaultOrchestratorConfig,
createEmptyAgentProgress,
parseTaskSpecFile,
parseSpawnResult,
MAX_TASK_FILE_SIZE,
MAX_CONTEXT_FILE_SIZE,
MAX_CONTEXT_FILES,
MAX_QUEUE_LENGTH,
BUDGET_WARNING_THRESHOLD,
MESSAGE_MAX_SIZE,
MAX_MESSAGES_PER_CHANNEL,
MAX_TRACKED_AGENTS,
} from './spawn-types.js';
import { generateAgentClaudeMd, buildInitialPrompt } from './spawn-claude-md.js';
import { getErrorMessage } from './types.js';
// ========== Local Constants ==========
/** UUID truncation length for fallback agent IDs */
const UUID_TRUNCATE_LENGTH = 8;
/** Message sequence number padding length */
const MESSAGE_SEQUENCE_PAD_LENGTH = 3;
/** Timeout warning threshold (90% of timeout) */
const TIMEOUT_WARNING_RATIO = 0.9;
/** Budget hard limit ratio (110% - force stop) */
const BUDGET_HARD_LIMIT_RATIO = 1.1;
/** Budget soft limit ratio (100% - warning) */
const BUDGET_SOFT_LIMIT_RATIO = 1.0;
// ========== Types for integration ==========
/**
* Interface for session creation callback.
* The orchestrator delegates session creation to the server to avoid circular deps.
*/
export interface SessionCreator {
createAgentSession(workingDir: string, name: string): Promise<{ sessionId: string }>;
writeToSession(sessionId: string, data: string): void;
getSessionTokens(sessionId: string): number;
getSessionCost(sessionId: string): number;
stopSession(sessionId: string): Promise<void>;
onSessionCompletion(sessionId: string, handler: (phrase: string) => void): void;
removeSessionCompletionHandler(sessionId: string, handler: (phrase: string) => void): void;
}
// ========== Events ==========
export interface SpawnOrchestratorEvents {
/** Agent added to queue */
queued: (data: { agentId: string; name: string; parentSessionId: string; position: number }) => void;
/** Agent directory being set up */
initializing: (data: { agentId: string; name: string; workingDir: string }) => void;
/** Agent session started */
started: (data: { agentId: string; name: string; sessionId: string }) => void;
/** Agent progress update */
progress: (data: { agentId: string; progress: AgentProgress }) => void;
/** New message in channel */
message: (data: { agentId: string; message: SpawnMessage }) => void;
/** Agent completed successfully */
completed: (data: { agentId: string; result: SpawnResult }) => void;
/** Agent failed */
failed: (data: { agentId: string; error: string; partialProgress: AgentProgress | null }) => void;
/** Agent timed out */
timeout: (data: { agentId: string; elapsed: number; limit: number }) => void;
/** Agent cancelled */
cancelled: (data: { agentId: string; reason: string }) => void;
/** Budget warning */
budgetWarning: (data: { agentId: string; type: 'tokens' | 'cost'; used: number; limit: number }) => void;
/** Overall state changed */
stateUpdate: (state: SpawnTrackerState) => void;
}
/**
* SpawnOrchestrator - Manages the full lifecycle of spawned agents.
*
* Handles agent creation, monitoring, communication, resource governance,
* and cleanup. Integrates with Session, ScreenManager, and RalphTracker.
*/
export class SpawnOrchestrator extends EventEmitter {
private _agents: Map<string, AgentContext> = new Map();
private _completedAgents: Map<string, AgentContext> = new Map();
private _queue: SpawnTask[] = [];
private _config: SpawnOrchestratorConfig;
private _sessionCreator: SessionCreator | null = null;
private _totalSpawned: number = 0;
private _totalCompleted: number = 0;
private _totalFailed: number = 0;
private _maxDepthReached: number = 0;
private _completionHandlers: Map<string, (phrase: string) => void> = new Map();
constructor(config?: Partial<SpawnOrchestratorConfig>) {
super();
this._config = { ...createDefaultOrchestratorConfig(), ...config };
}
/**
* Set the session creator callback.
* Must be called before any spawn requests can be processed.
*/
setSessionCreator(creator: SessionCreator): void {
this._sessionCreator = creator;
}
/**
* Get current orchestrator configuration.
*/
get config(): SpawnOrchestratorConfig {
return { ...this._config };
}
/**
* Update orchestrator configuration.
*/
updateConfig(config: Partial<SpawnOrchestratorConfig>): void {
Object.assign(this._config, config);
}
/**
* Handle a spawn request detected from terminal output.
*
* @param filePath - Path to the task spec file (relative to parent's workingDir)
* @param parentSessionId - ID of the parent session
* @param parentWorkingDir - Working directory of the parent session
* @param parentDepth - Depth of the parent in the spawn tree
*/
async handleSpawnRequest(
filePath: string,
parentSessionId: string,
parentWorkingDir: string,
parentDepth: number = 0
): Promise<void> {
if (!this._sessionCreator) {
console.error('[spawn-orchestrator] No session creator set, cannot spawn agent');
return;
}
// Resolve file path relative to parent's working directory
const resolvedPath = isAbsolute(filePath) ? filePath : join(parentWorkingDir, filePath);
// Validate file exists and size
if (!existsSync(resolvedPath)) {
console.error(`[spawn-orchestrator] Task file not found: ${resolvedPath}`);
this.emit('failed', { agentId: 'unknown', error: `Task file not found: ${resolvedPath}`, partialProgress: null });
return;
}
const stat = statSync(resolvedPath);
if (stat.size > MAX_TASK_FILE_SIZE) {
console.error(`[spawn-orchestrator] Task file too large: ${stat.size} bytes (max ${MAX_TASK_FILE_SIZE})`);
this.emit('failed', { agentId: 'unknown', error: `Task file too large: ${stat.size} bytes`, partialProgress: null });
return;
}
// Parse task file
const content = readFileSync(resolvedPath, 'utf-8');
const fallbackId = `agent-${uuidv4().slice(0, UUID_TRUNCATE_LENGTH)}`;
const parsed = parseTaskSpecFile(content, fallbackId);
if (!parsed) {
console.error(`[spawn-orchestrator] Failed to parse task file: ${resolvedPath}`);
this.emit('failed', { agentId: fallbackId, error: 'Failed to parse task spec YAML frontmatter', partialProgress: null });
return;
}
const childDepth = parentDepth + 1;
// Depth check
if (childDepth > this._config.maxSpawnDepth) {
console.error(`[spawn-orchestrator] Max spawn depth (${this._config.maxSpawnDepth}) exceeded at depth ${childDepth}`);
this.emit('failed', { agentId: parsed.spec.agentId, error: `Max spawn depth exceeded (${this._config.maxSpawnDepth})`, partialProgress: null });
return;
}
// Enforce timeout limits
if (parsed.spec.timeoutMinutes > this._config.maxTimeoutMinutes) {
parsed.spec.timeoutMinutes = this._config.maxTimeoutMinutes;
}
const task: SpawnTask = {
spec: parsed.spec,
instructions: parsed.instructions,
sourceFile: resolvedPath,
parentSessionId,
depth: childDepth,
};
// Check dependencies
if (task.spec.dependsOn && task.spec.dependsOn.length > 0) {
const unmetDeps = task.spec.dependsOn.filter(depId => {
const dep = this._completedAgents.get(depId);
return !dep || dep.status !== 'completed';
});
if (unmetDeps.length > 0) {
// Queue with dependency tracking
this.enqueueTask(task);
return;
}
}
// Concurrency check
const activeCount = this.getActiveCount();
if (activeCount >= this._config.maxConcurrentAgents) {
this.enqueueTask(task);
return;
}
// Spawn immediately
await this.spawnAgent(task);
}
/**
* Cancel an agent by ID.
* Cascades cancellation to all child agents before cleaning up the parent.
*/
async cancelAgent(agentId: string, reason: string = 'Cancelled by parent'): Promise<void> {
const agent = this._agents.get(agentId);
if (!agent) {
// Check queue
const queueIdx = this._queue.findIndex(t => t.spec.agentId === agentId);
if (queueIdx >= 0) {
this._queue.splice(queueIdx, 1);
this.emit('cancelled', { agentId, reason: 'Removed from queue' });
this.emitStateUpdate();
}
return;
}
// Cancel all child agents first (cascade)
// Child agents have their parentSessionId set to this agent's sessionId
if (agent.sessionId) {
for (const [childId, childAgent] of this._agents) {
if (childAgent.parentSessionId === agent.sessionId && childAgent.status !== 'cancelled') {
await this.cancelAgent(childId, `Parent ${agentId} cancelled`);
}
}
}
// Also remove any queued tasks that depend on this agent's session
if (agent.sessionId) {
const queuedChildren = this._queue.filter(t => t.parentSessionId === agent.sessionId);
for (const task of queuedChildren) {
const idx = this._queue.indexOf(task);
if (idx >= 0) {
this._queue.splice(idx, 1);
this.emit('cancelled', { agentId: task.spec.agentId, reason: `Parent ${agentId} cancelled` });
}
}
}
agent.status = 'cancelled';
this.emit('cancelled', { agentId, reason });
await this.cleanupAgent(agentId);
}
/**
* Send a message to an agent.
*/
async sendMessageToAgent(agentId: string, content: string): Promise<void> {
const agent = this._agents.get(agentId);
if (!agent) return;
if (content.length > MESSAGE_MAX_SIZE) {
content = content.slice(0, MESSAGE_MAX_SIZE);
}
const messagesDir = join(agent.commsDir, 'messages');
if (!existsSync(messagesDir)) {
mkdirSync(messagesDir, { recursive: true });
}
// Count existing messages
const existingMessages = readdirSync(messagesDir).filter(f => f.endsWith('.md'));
if (existingMessages.length >= MAX_MESSAGES_PER_CHANNEL) {
return; // Channel full
}
const seq = existingMessages.length + 1;
const seqStr = String(seq).padStart(MESSAGE_SEQUENCE_PAD_LENGTH, '0');
const fileName = `${seqStr}-parent.md`;
const message: SpawnMessage = {
sequence: seq,
sender: 'parent',
content,
sentAt: Date.now(),
read: false,
};
writeFileSync(join(messagesDir, fileName), content, 'utf-8');
this.emit('message', { agentId, message });
}
/**
* Get status of a specific agent.
*/
getAgentStatus(agentId: string): AgentStatusReport | null {
const agent = this._agents.get(agentId) || this._completedAgents.get(agentId);
if (!agent) return null;
return this.buildStatusReport(agent);
}
/**
* Get status of all agents (active + recently completed).
*/
getAllAgentStatuses(): AgentStatusReport[] {
const reports: AgentStatusReport[] = [];
for (const agent of this._agents.values()) {
reports.push(this.buildStatusReport(agent));
}
for (const agent of this._completedAgents.values()) {
reports.push(this.buildStatusReport(agent));
}
return reports;
}
/**
* Get current orchestrator state.
*/
getState(): SpawnTrackerState {
return {
enabled: true,
activeCount: this.getActiveCount(),
queuedCount: this._queue.length,
totalSpawned: this._totalSpawned,
totalCompleted: this._totalCompleted,
totalFailed: this._totalFailed,
maxDepthReached: this._maxDepthReached,
agents: this.getAllAgentStatuses(),
};
}
/**
* Get state for persistence.
*/
getPersistedState(): SpawnPersistedState {
const agents: SpawnPersistedState['agents'] = {};
for (const [id, agent] of this._agents) {
agents[id] = {
agentId: id,
status: agent.status,
parentSessionId: agent.parentSessionId,
childSessionId: agent.sessionId,
depth: agent.depth,
startedAt: agent.startedAt,
commsDir: agent.commsDir,
workingDir: agent.workingDir,
completionPhrase: agent.task.spec.completionPhrase,
timeoutMinutes: agent.task.spec.timeoutMinutes,
};
}
return { config: this._config, agents };
}
/**
* Stop all agents.
*/
async stopAll(): Promise<void> {
const agentIds = Array.from(this._agents.keys());
for (const agentId of agentIds) {
await this.cancelAgent(agentId, 'Orchestrator shutdown');
}
this._queue = [];
}
/**
* Read an agent's result.md file.
*/
readAgentResult(agentId: string): SpawnResult | null {
const agent = this._agents.get(agentId) || this._completedAgents.get(agentId);
if (!agent) return null;
const resultPath = join(agent.commsDir, 'result.md');
if (!existsSync(resultPath)) return null;
const content = readFileSync(resultPath, 'utf-8');
const durationMs = agent.startedAt ? Date.now() - agent.startedAt : 0;
return parseSpawnResult(content, agentId, durationMs);
}
/**
* Read an agent's progress.json file.
*/
readAgentProgress(agentId: string): AgentProgress | null {
const agent = this._agents.get(agentId) || this._completedAgents.get(agentId);
if (!agent) return null;
const progressPath = join(agent.commsDir, 'progress.json');
if (!existsSync(progressPath)) return null;
try {
const content = readFileSync(progressPath, 'utf-8');
return JSON.parse(content) as AgentProgress;
} catch (err) {
console.warn(`[spawn-orchestrator] Failed to read progress for agent ${agentId}: ${getErrorMessage(err)}`);
return null;
}
}
/**
* Read messages from an agent's communication channel.
*/
readAgentMessages(agentId: string): SpawnMessage[] {
const agent = this._agents.get(agentId) || this._completedAgents.get(agentId);
if (!agent) return [];
const messagesDir = join(agent.commsDir, 'messages');
if (!existsSync(messagesDir)) return [];
try {
const files = readdirSync(messagesDir)
.filter(f => f.endsWith('.md'))
.sort();
const messages: SpawnMessage[] = [];
for (const file of files) {
const match = file.match(/^(\d+)-(parent|agent)\.md$/);
if (!match) continue;
try {
const filePath = join(messagesDir, file);
const content = readFileSync(filePath, 'utf-8');
messages.push({
sequence: parseInt(match[1]),
sender: match[2] as 'parent' | 'agent',
content,
sentAt: statSync(filePath).mtimeMs,
read: true,
});
} catch (err) {
console.warn(`[spawn-orchestrator] Failed to read message file ${file}: ${getErrorMessage(err)}`);
// Continue processing other messages
}
}
return messages;
} catch (err) {
console.warn(`[spawn-orchestrator] Failed to read messages for agent ${agentId}: ${getErrorMessage(err)}`);
return [];
}
}
/**
* Programmatically trigger a spawn without terminal detection.
*/
async triggerSpawn(
taskContent: string,
parentSessionId: string,
parentWorkingDir: string,
parentDepth: number = 0
): Promise<string | null> {
const fallbackId = `agent-${uuidv4().slice(0, UUID_TRUNCATE_LENGTH)}`;
const parsed = parseTaskSpecFile(taskContent, fallbackId);
if (!parsed) return null;
// If the spec doesn't specify a workingDir, use the parent's
if (!parsed.spec.workingDir) {
parsed.spec.workingDir = parentWorkingDir;
}
// Write task content to a temp file so setupAgentDirectory can read it
const tempDir = join(this._config.casesDir, '.spawn-tmp');
mkdirSync(tempDir, { recursive: true });
const tempFile = join(tempDir, `${parsed.spec.agentId}.md`);
writeFileSync(tempFile, taskContent, 'utf-8');
const task: SpawnTask = {
spec: parsed.spec,
instructions: parsed.instructions,
sourceFile: tempFile,
parentSessionId,
depth: parentDepth + 1,
};
await this.spawnAgent(task);
return task.spec.agentId;
}
// ========== Internal Methods ==========
private getActiveCount(): number {
let count = 0;
for (const agent of this._agents.values()) {
if (agent.status === 'initializing' || agent.status === 'running') {
count++;
}
}
return count;
}
private enqueueTask(task: SpawnTask): void {
if (this._queue.length >= MAX_QUEUE_LENGTH) {
this.emit('failed', {
agentId: task.spec.agentId,
error: `Queue full (max ${MAX_QUEUE_LENGTH})`,
partialProgress: null,
});
return;
}
// Insert by priority (higher priority first)
const priorityOrder = { critical: 0, high: 1, normal: 2, low: 3 };
const taskPriority = priorityOrder[task.spec.priority];
let insertIdx = this._queue.length;
for (let i = 0; i < this._queue.length; i++) {
if (priorityOrder[this._queue[i].spec.priority] > taskPriority) {
insertIdx = i;
break;
}
}
this._queue.splice(insertIdx, 0, task);
this.emit('queued', {
agentId: task.spec.agentId,
name: task.spec.name,
parentSessionId: task.parentSessionId,
position: insertIdx + 1,
});
this.emitStateUpdate();
}
private async spawnAgent(task: SpawnTask): Promise<void> {
if (!this._sessionCreator) return;
const agentId = task.spec.agentId;
this._totalSpawned++;
if (task.depth > this._maxDepthReached) {
this._maxDepthReached = task.depth;
}
// Create agent context
const workingDir = join(this._config.casesDir, `spawn-${agentId}`);
const commsDir = join(workingDir, 'spawn-comms');
const agent: AgentContext = {
task,
sessionId: null,
workingDir,
commsDir,
parentSessionId: task.parentSessionId,
depth: task.depth,
timeoutTimer: null,
warningTimer: null,
progressTimer: null,
status: 'initializing',
startedAt: null,
tokenBudget: task.spec.maxTokens ?? null,
costBudget: task.spec.maxCost ?? null,
};
this._agents.set(agentId, agent);
this.emit('initializing', { agentId, name: task.spec.name, workingDir });
this.emitStateUpdate();
try {
// Setup directory structure
this.setupAgentDirectory(task, workingDir, commsDir);
// Create session
const { sessionId } = await this._sessionCreator.createAgentSession(workingDir, agentId);
agent.sessionId = sessionId;
agent.status = 'running';
agent.startedAt = Date.now();
this.emit('started', { agentId, name: task.spec.name, sessionId });
this.emitStateUpdate();
// Setup completion listener
this.setupCompletionListener(agent);
// Setup progress monitor
this.setupProgressMonitor(agent);
// Setup timeout
this.setupTimeout(agent);
// Inject initial prompt (short delay to let session initialize)
setTimeout(() => {
if (agent.status === 'running' && this._sessionCreator) {
const prompt = buildInitialPrompt(task);
this._sessionCreator.writeToSession(sessionId, prompt + '\r');
}
}, 3000);
} catch (err) {
agent.status = 'failed';
this._totalFailed++;
this.emit('failed', { agentId, error: getErrorMessage(err), partialProgress: null });
await this.cleanupAgent(agentId);
}
}
private setupAgentDirectory(task: SpawnTask, workingDir: string, commsDir: string): void {
// Create directory structure
mkdirSync(workingDir, { recursive: true });
mkdirSync(commsDir, { recursive: true });
mkdirSync(join(commsDir, 'messages'), { recursive: true });
mkdirSync(join(commsDir, 'artifacts'), { recursive: true });
mkdirSync(join(workingDir, 'workspace'), { recursive: true });
// Copy task.md to comms
writeFileSync(join(commsDir, 'task.md'), readFileSync(task.sourceFile, 'utf-8'), 'utf-8');
// Write initial progress.json
writeFileSync(
join(commsDir, 'progress.json'),
JSON.stringify(createEmptyAgentProgress(), null, 2),
'utf-8'
);
// Generate and write CLAUDE.md
const claudeMd = generateAgentClaudeMd(task, commsDir, workingDir);
writeFileSync(join(workingDir, 'CLAUDE.md'), claudeMd, 'utf-8');
// Symlink context files into workspace
if (task.spec.contextFiles && task.spec.contextFiles.length > 0) {
const parentWorkingDir = this.resolveParentWorkingDir(task);
let fileCount = 0;
for (const contextFile of task.spec.contextFiles) {
if (fileCount >= MAX_CONTEXT_FILES) break;
const sourcePath = isAbsolute(contextFile)
? contextFile
: join(parentWorkingDir, contextFile);
if (!existsSync(sourcePath)) continue;
const stat = statSync(sourcePath);
if (stat.size > MAX_CONTEXT_FILE_SIZE) continue;
const destPath = join(workingDir, 'workspace', contextFile.split('/').pop() || contextFile);
try {
symlinkSync(sourcePath, destPath);
fileCount++;
} catch {
// Ignore symlink errors (e.g., dest already exists)
}
}
}
}
private resolveParentWorkingDir(task: SpawnTask): string {
// If the task has a specified workingDir, resolve it
if (task.spec.workingDir) {
return isAbsolute(task.spec.workingDir)
? task.spec.workingDir
: resolve(this._config.casesDir, task.spec.workingDir);
}
// Default: use casesDir
return this._config.casesDir;
}
private setupCompletionListener(agent: AgentContext): void {
if (!this._sessionCreator || !agent.sessionId) return;
const handler = (phrase: string) => {
if (phrase === agent.task.spec.completionPhrase) {
this.handleAgentCompletion(agent);
}
};
this._completionHandlers.set(agent.task.spec.agentId, handler);
this._sessionCreator.onSessionCompletion(agent.sessionId, handler);
}
private setupProgressMonitor(agent: AgentContext): void {
if (this._config.progressPollIntervalMs <= 0) return;
agent.progressTimer = setInterval(() => {
if (agent.status !== 'running') return;
// Read progress
const progress = this.readAgentProgress(agent.task.spec.agentId);
if (progress) {
this.emit('progress', { agentId: agent.task.spec.agentId, progress });
}
// Check resource budgets
this.checkResourceBudgets(agent);
}, this._config.progressPollIntervalMs);
}
private setupTimeout(agent: AgentContext): void {
const timeoutMs = agent.task.spec.timeoutMinutes * 60 * 1000;
// Warning at 90% - store timer for cleanup
const warningMs = timeoutMs * TIMEOUT_WARNING_RATIO;
agent.warningTimer = setTimeout(() => {
if (agent.status === 'running' && this._sessionCreator && agent.sessionId) {
this._sessionCreator.writeToSession(
agent.sessionId,
'WARNING: You have less than 10% of your timeout remaining. Please wrap up and write your result.md soon.\r'
);
}
}, warningMs);
// Hard timeout
agent.timeoutTimer = setTimeout(() => {
if (agent.status === 'running') {
this.handleAgentTimeout(agent);
}
}, timeoutMs);
}
private checkResourceBudgets(agent: AgentContext): void {
if (!this._sessionCreator || !agent.sessionId) return;
// Token budget
if (agent.tokenBudget !== null) {
const tokensUsed = this._sessionCreator.getSessionTokens(agent.sessionId);
const ratio = tokensUsed / agent.tokenBudget;
if (ratio >= BUDGET_HARD_LIMIT_RATIO) {
// Force kill at 110%
this.handleAgentTimeout(agent);
return;
} else if (ratio >= BUDGET_SOFT_LIMIT_RATIO) {
// Graceful shutdown
this._sessionCreator.writeToSession(
agent.sessionId,
'You have exceeded your token budget. Write your result.md NOW and output your completion phrase.\r'
);
} else if (ratio >= BUDGET_WARNING_THRESHOLD) {
this.emit('budgetWarning', {
agentId: agent.task.spec.agentId,
type: 'tokens',
used: tokensUsed,
limit: agent.tokenBudget,
});
}
}
// Cost budget
if (agent.costBudget !== null) {
const costUsed = this._sessionCreator.getSessionCost(agent.sessionId);
const ratio = costUsed / agent.costBudget;
if (ratio >= BUDGET_HARD_LIMIT_RATIO) {
this.handleAgentTimeout(agent);
return;
} else if (ratio >= BUDGET_SOFT_LIMIT_RATIO) {
this._sessionCreator.writeToSession(
agent.sessionId,
'You have exceeded your cost budget. Write your result.md NOW and output your completion phrase.\r'
);
} else if (ratio >= BUDGET_WARNING_THRESHOLD) {
this.emit('budgetWarning', {
agentId: agent.task.spec.agentId,
type: 'cost',
used: costUsed,
limit: agent.costBudget,
});
}
}
}
private async handleAgentCompletion(agent: AgentContext): Promise<void> {
if (agent.status !== 'running') return;
agent.status = 'completing';
this._totalCompleted++;
// Read result
const result = this.readAgentResult(agent.task.spec.agentId);
if (result) {
// Update token/cost from session
if (this._sessionCreator && agent.sessionId) {
result.tokens.total = this._sessionCreator.getSessionTokens(agent.sessionId);
result.cost = this._sessionCreator.getSessionCost(agent.sessionId);
}
this.emit('completed', { agentId: agent.task.spec.agentId, result });
} else {
// No result file found, create a minimal one
const minimalResult: SpawnResult = {
status: 'completed',
durationMs: agent.startedAt ? Date.now() - agent.startedAt : 0,
tokens: { input: 0, output: 0, total: 0 },
cost: 0,
summary: 'Agent completed but no result.md was found',
output: '',
filesChanged: [],
agentId: agent.task.spec.agentId,
completedAt: Date.now(),
};
this.emit('completed', { agentId: agent.task.spec.agentId, result: minimalResult });
}
agent.status = 'completed';
await this.cleanupAgent(agent.task.spec.agentId);
this.processQueue();
}
private async handleAgentTimeout(agent: AgentContext): Promise<void> {
if (agent.status !== 'running') return;
agent.status = 'timeout';
this._totalFailed++;
const elapsed = agent.startedAt ? Date.now() - agent.startedAt : 0;
const limit = agent.task.spec.timeoutMinutes * 60 * 1000;
this.emit('timeout', { agentId: agent.task.spec.agentId, elapsed, limit });
await this.cleanupAgent(agent.task.spec.agentId);
this.processQueue();
}
private async cleanupAgent(agentId: string): Promise<void> {
const agent = this._agents.get(agentId);
if (!agent) return;
// Clear timers
if (agent.timeoutTimer) {
clearTimeout(agent.timeoutTimer);
agent.timeoutTimer = null;
}
if (agent.warningTimer) {
clearTimeout(agent.warningTimer);
agent.warningTimer = null;
}
if (agent.progressTimer) {
clearInterval(agent.progressTimer);
agent.progressTimer = null;
}
// Remove completion handler
const handler = this._completionHandlers.get(agentId);
if (handler && this._sessionCreator && agent.sessionId) {
this._sessionCreator.removeSessionCompletionHandler(agent.sessionId, handler);
this._completionHandlers.delete(agentId);
}
// Stop session
if (agent.sessionId && this._sessionCreator) {
try {
await this._sessionCreator.stopSession(agent.sessionId);
} catch (err) {
console.warn(`[spawn-orchestrator] Failed to stop session for agent ${agentId}: ${getErrorMessage(err)}`);
}
}
// Move to completed (LRU)
this._agents.delete(agentId);
this._completedAgents.set(agentId, agent);
// LRU eviction for completed agents
if (this._completedAgents.size > MAX_TRACKED_AGENTS) {
const firstKey = this._completedAgents.keys().next().value;
if (firstKey) this._completedAgents.delete(firstKey);
}
this.emitStateUpdate();
}
private processQueue(): void {
while (this._queue.length > 0 && this.getActiveCount() < this._config.maxConcurrentAgents) {
const task = this._queue.shift();
if (!task) break;
// Re-check dependencies
if (task.spec.dependsOn && task.spec.dependsOn.length > 0) {
const unmetDeps = task.spec.dependsOn.filter(depId => {
const dep = this._completedAgents.get(depId);
return !dep || dep.status !== 'completed';
});
if (unmetDeps.length > 0) {
// Put back in queue
this._queue.unshift(task);
break;
}
}
// Spawn (async, don't await to allow multiple spawns)
this.spawnAgent(task).catch(err => {
console.error(`[spawn-orchestrator] Failed to spawn queued agent: ${getErrorMessage(err)}`);
});
}
}
private buildStatusReport(agent: AgentContext): AgentStatusReport {
const now = Date.now();
const elapsed = agent.startedAt ? now - agent.startedAt : 0;
const timeoutMs = agent.task.spec.timeoutMinutes * 60 * 1000;
const timeRemaining = agent.startedAt ? Math.max(0, timeoutMs - elapsed) : timeoutMs;
let tokensUsed = 0;
let costSoFar = 0;
if (agent.sessionId && this._sessionCreator) {
tokensUsed = this._sessionCreator.getSessionTokens(agent.sessionId);
costSoFar = this._sessionCreator.getSessionCost(agent.sessionId);
}
// Check dependency status
let dependencyStatus: 'waiting' | 'ready' | 'n/a' = 'n/a';
if (agent.task.spec.dependsOn && agent.task.spec.dependsOn.length > 0) {
const allMet = agent.task.spec.dependsOn.every(depId => {
const dep = this._completedAgents.get(depId);
return dep && dep.status === 'completed';
});
dependencyStatus = allMet ? 'ready' : 'waiting';
}
return {
agentId: agent.task.spec.agentId,
name: agent.task.spec.name,
type: agent.task.spec.type,
status: agent.status,
priority: agent.task.spec.priority,
parentSessionId: agent.parentSessionId,
childSessionId: agent.sessionId,
depth: agent.depth,
startedAt: agent.startedAt,
elapsedMs: elapsed,
progress: this.readAgentProgress(agent.task.spec.agentId),
tokensUsed,
costSoFar,
tokenBudget: agent.tokenBudget,
costBudget: agent.costBudget,
timeoutMinutes: agent.task.spec.timeoutMinutes,
timeRemainingMs: timeRemaining,
completionPhrase: agent.task.spec.completionPhrase,
dependsOn: agent.task.spec.dependsOn || [],
dependencyStatus,
};
}
private emitStateUpdate(): void {
this.emit('stateUpdate', this.getState());
}
}
-18
View File
@@ -1255,24 +1255,6 @@ export interface ImageDetectedEvent {
size: number;
}
// ========== Spawn1337 Protocol Re-exports ==========
export type {
SpawnPriority,
SpawnResultDelivery,
SpawnStatus,
SpawnTaskSpec,
SpawnTask,
AgentProgress,
SpawnResult,
SpawnMessage,
AgentStatusReport,
SpawnTrackerState,
SpawnOrchestratorConfig,
AgentContext,
SpawnPersistedState,
} from './spawn-types.js';
// ========== Execution Bridge Re-exports ==========
export type {
+25 -51
View File
@@ -466,7 +466,7 @@ class ClaudemanApp {
this._subagentHideTimeout = null; // Timeout for hover-based dropdown hide
this.ralphStatePanelCollapsed = true; // Default to collapsed
// Plan subagent windows (visible Opus agents during plan generation)
// Plan subagent windows (visible agents during plan generation)
this.planSubagents = new Map(); // Map<agentId, { type, model, status, startTime, element, relativePos }>
this.planSubagentWindowZIndex = 1100;
this.planGenerationStopped = false; // Flag to ignore SSE events after Stop
@@ -641,7 +641,7 @@ class ClaudemanApp {
fontFamily: '"Fira Code", "Cascadia Code", "JetBrains Mono", "SF Mono", Monaco, monospace',
fontSize: 14,
lineHeight: 1.2,
cursorBlink: true,
cursorBlink: false,
cursorStyle: 'block',
scrollback: scrollback,
allowTransparency: true,
@@ -1146,6 +1146,10 @@ class ClaudemanApp {
// This connects subagents that were waiting for the session to identify itself
if (claudeSessionIdJustSet) {
this.recheckOrphanSubagents();
// Update connection lines after DOM settles (ensure tabs are rendered)
requestAnimationFrame(() => {
this.updateConnectionLines();
});
}
});
@@ -1690,55 +1694,6 @@ class ClaudemanApp {
this.handleBashToolsUpdate(data.sessionId, data.tools);
});
// Spawn agent notification events
addListener('spawn:failed', (e) => {
const data = JSON.parse(e.data);
this.notificationManager?.notify({
urgency: 'critical',
category: 'spawn-failed',
sessionId: data.sessionId,
sessionName: data.agentId || 'agent',
title: 'Agent Failed',
message: `Agent "${data.agentId}" failed: ${data.reason || 'unknown'}`,
});
});
addListener('spawn:timeout', (e) => {
const data = JSON.parse(e.data);
this.notificationManager?.notify({
urgency: 'critical',
category: 'spawn-timeout',
sessionId: data.sessionId,
sessionName: data.agentId || 'agent',
title: 'Agent Timeout',
message: `Agent "${data.agentId}" exceeded time limit`,
});
});
addListener('spawn:budgetWarning', (e) => {
const data = JSON.parse(e.data);
this.notificationManager?.notify({
urgency: 'warning',
category: 'spawn-budget',
sessionId: data.sessionId,
sessionName: data.agentId || 'agent',
title: 'Budget Warning',
message: `Agent "${data.agentId}" at ${data.percent || 80}% budget`,
});
});
addListener('spawn:completed', (e) => {
const data = JSON.parse(e.data);
this.notificationManager?.notify({
urgency: 'info',
category: 'spawn-completed',
sessionId: data.sessionId,
sessionName: data.agentId || 'agent',
title: 'Agent Complete',
message: `Agent "${data.agentId}" finished successfully`,
});
});
// Hook events (from Claude Code hooks system)
// Use pendingHooks state machine to track hook events and derive tab alerts.
// This ensures alerts persist even when session:working events fire.
@@ -1832,6 +1787,11 @@ class ClaudemanApp {
if (data.status === 'active') {
this.openSubagentWindow(data.agentId);
}
// Ensure connection lines are updated after window is created and DOM settles
requestAnimationFrame(() => {
this.updateConnectionLines();
});
});
addListener('subagent:updated', (e) => {
@@ -7658,6 +7618,11 @@ class ClaudemanApp {
this.renderSessionTabs(); // Update tab badges
this.saveSubagentWindowStates(); // Persist corrected mappings
// Update connection lines after all windows are restored (use rAF to ensure DOM is ready)
requestAnimationFrame(() => {
this.updateConnectionLines();
});
}
// ========== Help Modal ==========
@@ -9425,11 +9390,20 @@ class ClaudemanApp {
* Called when session:updated fires, in case claudeSessionId was just set.
*/
recheckOrphanSubagents() {
let anyFound = false;
for (const [agentId, agent] of this.subagents) {
if (!agent.parentSessionId && agent.sessionId) {
const hadParent = agent.parentSessionId;
this.findParentSessionForSubagent(agentId);
if (!hadParent && this.subagents.get(agentId)?.parentSessionId) {
anyFound = true;
}
}
}
// Ensure connection lines are updated after all orphans are processed
if (anyFound) {
this.updateConnectionLines();
}
}
/**
-207
View File
@@ -22,8 +22,6 @@ import { EventEmitter } from 'node:events';
import { Session, ClaudeMessage, type BackgroundTask, type RalphTrackerState, type RalphTodoItem, type ActiveBashTool } from '../session.js';
import { fileStreamManager } from '../file-stream-manager.js';
import { RespawnController, RespawnConfig, RespawnState } from '../respawn-controller.js';
import { SpawnOrchestrator, type SessionCreator } from '../spawn-orchestrator.js';
import type { SpawnOrchestratorConfig } from '../spawn-types.js';
import { ScreenManager } from '../screen-manager.js';
import { getStore } from '../state-store.js';
import { generateClaudeMd } from '../templates/claude-md.js';
@@ -307,8 +305,6 @@ export class WebServer extends EventEmitter {
private sseHealthCheckTimer: NodeJS.Timeout | null = null;
// Flag to prevent new timers during shutdown
private _isStopping: boolean = false;
// Spawn1337 agent orchestrator
private spawnOrchestrator: SpawnOrchestrator;
// Token recording for daily stats (track what's been recorded to avoid double-counting)
private lastRecordedTokens: Map<string, { input: number; output: number }> = new Map();
private tokenRecordingTimer: NodeJS.Timeout | null = null;
@@ -350,10 +346,6 @@ export class WebServer extends EventEmitter {
this.broadcast('screen:statsUpdated', screens);
});
// Initialize spawn orchestrator
this.spawnOrchestrator = new SpawnOrchestrator();
this.setupSpawnOrchestratorListeners();
// Initialize execution bridge with model config from settings
this.executionBridge = getExecutionBridge(this.loadModelConfig());
this.setupExecutionBridgeListeners();
@@ -510,7 +502,6 @@ export class WebServer extends EventEmitter {
// Returns comprehensive memory metrics for debugging memory leaks
this.app.get('/api/debug/memory', async () => {
const mem = process.memoryUsage();
const spawnState = this.spawnOrchestrator.getState();
const subagentStats = subagentWatcher.getStats();
// Calculate total Map entries for memory estimation
@@ -569,13 +560,6 @@ export class WebServer extends EventEmitter {
subagentIdleTimers: subagentStats.idleTimerCount,
total: this.respawnTimers.size + this.pendingRespawnStarts.size + subagentStats.idleTimerCount,
},
spawn: {
activeAgents: spawnState.activeCount,
queuedAgents: spawnState.queuedCount,
totalSpawned: spawnState.totalSpawned,
totalCompleted: spawnState.totalCompleted,
totalFailed: spawnState.totalFailed,
},
uptime: {
seconds: Math.round(process.uptime()),
formatted: formatUptime(process.uptime()),
@@ -1987,9 +1971,6 @@ export class WebServer extends EventEmitter {
const claudeMd = generateClaudeMd(name, description || '', templatePath);
writeFileSync(join(casePath, 'CLAUDE.md'), claudeMd);
// Write .mcp.json for Claude Code to discover spawn tools
this.writeMcpConfig(casePath);
// Write .claude/settings.local.json with hooks for desktop notifications
writeHooksConfig(casePath);
@@ -2236,9 +2217,6 @@ export class WebServer extends EventEmitter {
const claudeMd = generateClaudeMd(caseName, '', templatePath);
writeFileSync(join(casePath, 'CLAUDE.md'), claudeMd);
// Write .mcp.json for Claude Code to discover spawn tools
this.writeMcpConfig(casePath);
// Write .claude/settings.local.json with hooks for desktop notifications
writeHooksConfig(casePath);
@@ -3019,93 +2997,6 @@ NOW: Generate the implementation plan for the task above. Think step by step.`;
return this.getSystemStats();
});
// ========== Spawn1337 Agent Protocol Endpoints ==========
this.app.get('/api/spawn/agents', async () => {
return { success: true, data: this.spawnOrchestrator.getAllAgentStatuses() };
});
this.app.get('/api/spawn/agents/:agentId', async (req) => {
const { agentId } = req.params as { agentId: string };
const status = this.spawnOrchestrator.getAgentStatus(agentId);
if (!status) {
return createErrorResponse(ApiErrorCode.NOT_FOUND, `Agent ${agentId} not found`);
}
return { success: true, data: status };
});
this.app.get('/api/spawn/agents/:agentId/result', async (req) => {
const { agentId } = req.params as { agentId: string };
const result = this.spawnOrchestrator.readAgentResult(agentId);
if (!result) {
return createErrorResponse(ApiErrorCode.NOT_FOUND, `No result found for agent ${agentId}`);
}
return { success: true, data: result };
});
this.app.get('/api/spawn/agents/:agentId/progress', async (req) => {
const { agentId } = req.params as { agentId: string };
const progress = this.spawnOrchestrator.readAgentProgress(agentId);
return { success: true, data: progress };
});
this.app.get('/api/spawn/agents/:agentId/messages', async (req) => {
const { agentId } = req.params as { agentId: string };
const messages = this.spawnOrchestrator.readAgentMessages(agentId);
return { success: true, data: messages };
});
this.app.post('/api/spawn/agents/:agentId/message', async (req) => {
const { agentId } = req.params as { agentId: string };
const { content } = req.body as { content: string };
if (!content) {
return createErrorResponse(ApiErrorCode.INVALID_INPUT, 'Message content is required');
}
await this.spawnOrchestrator.sendMessageToAgent(agentId, content);
return { success: true };
});
this.app.post('/api/spawn/agents/:agentId/cancel', async (req) => {
const { agentId } = req.params as { agentId: string };
const { reason } = (req.body as { reason?: string }) || {};
await this.spawnOrchestrator.cancelAgent(agentId, reason || 'Cancelled via API');
return { success: true };
});
this.app.delete('/api/spawn/agents/:agentId', async (req) => {
const { agentId } = req.params as { agentId: string };
await this.spawnOrchestrator.cancelAgent(agentId, 'Force killed via API');
return { success: true };
});
this.app.get('/api/spawn/status', async () => {
return { success: true, data: this.spawnOrchestrator.getState() };
});
this.app.put('/api/spawn/config', async (req) => {
const config = req.body as Partial<SpawnOrchestratorConfig>;
this.spawnOrchestrator.updateConfig(config);
return { success: true, data: this.spawnOrchestrator.config };
});
this.app.post('/api/spawn/trigger', async (req) => {
const { taskContent, parentSessionId, parentWorkingDir } = req.body as {
taskContent: string;
parentSessionId: string;
parentWorkingDir?: string;
};
if (!taskContent || !parentSessionId) {
return createErrorResponse(ApiErrorCode.INVALID_INPUT, 'taskContent and parentSessionId are required');
}
const session = this.sessions.get(parentSessionId);
const workingDir = parentWorkingDir || session?.workingDir || process.cwd();
const agentId = await this.spawnOrchestrator.triggerSpawn(taskContent, parentSessionId, workingDir);
if (!agentId) {
return createErrorResponse(ApiErrorCode.OPERATION_FAILED, 'Failed to parse task spec');
}
return { success: true, data: { agentId } };
});
// ========== Execution Bridge Endpoints ==========
// Get execution status
@@ -3971,83 +3862,6 @@ NOW: Generate the implementation plan for the task above. Think step by step.`;
});
}
private setupSpawnOrchestratorListeners(): void {
const sessionCreator: SessionCreator = {
createAgentSession: async (workingDir: string, name: string) => {
const globalNice = this.getGlobalNiceConfig();
const session = new Session({
workingDir,
screenManager: this.screenManager,
useScreen: true,
mode: 'claude',
name: `spawn:${name}`,
niceConfig: globalNice,
});
this.sessions.set(session.id, session);
this.store.incrementSessionsCreated();
this.setupSessionListeners(session);
session.parentAgentId = name;
await session.startInteractive();
this.broadcast('session:created', session.toDetailedState());
this.broadcast('session:interactive', { id: session.id });
this.persistSessionState(session);
// Configure ralph tracker for completion detection
session.ralphTracker.enable();
return { sessionId: session.id };
},
writeToSession: (sessionId: string, data: string) => {
const session = this.sessions.get(sessionId);
if (session) {
session.writeViaScreen(data);
}
},
getSessionTokens: (sessionId: string) => {
const session = this.sessions.get(sessionId);
return session ? session.totalTokens : 0;
},
getSessionCost: (sessionId: string) => {
const session = this.sessions.get(sessionId);
return session ? session.totalCost : 0;
},
stopSession: async (sessionId: string) => {
// Use cleanupSession to properly clean up all resources (respawn controllers,
// run summary trackers, file streams, Ralph state, etc.)
await this.cleanupSession(sessionId);
},
onSessionCompletion: (sessionId: string, handler: (phrase: string) => void) => {
const session = this.sessions.get(sessionId);
if (session) {
session.on('ralphCompletionDetected', handler);
}
},
removeSessionCompletionHandler: (sessionId: string, handler: (phrase: string) => void) => {
const session = this.sessions.get(sessionId);
if (session) {
session.off('ralphCompletionDetected', handler);
}
},
};
this.spawnOrchestrator.setSessionCreator(sessionCreator);
// Forward orchestrator events as SSE broadcasts
this.spawnOrchestrator.on('queued', (data) => this.broadcast('spawn:queued', data));
this.spawnOrchestrator.on('initializing', (data) => this.broadcast('spawn:initializing', data));
this.spawnOrchestrator.on('started', (data) => this.broadcast('spawn:started', data));
this.spawnOrchestrator.on('progress', (data) => this.broadcast('spawn:progress', data));
this.spawnOrchestrator.on('message', (data) => this.broadcast('spawn:message', data));
this.spawnOrchestrator.on('completed', (data) => this.broadcast('spawn:completed', data));
this.spawnOrchestrator.on('failed', (data) => this.broadcast('spawn:failed', data));
this.spawnOrchestrator.on('timeout', (data) => this.broadcast('spawn:timeout', data));
this.spawnOrchestrator.on('cancelled', (data) => this.broadcast('spawn:cancelled', data));
this.spawnOrchestrator.on('budgetWarning', (data) => this.broadcast('spawn:budgetWarning', data));
this.spawnOrchestrator.on('stateUpdate', (data) => this.broadcast('spawn:stateUpdate', data));
}
/**
* Load model configuration from settings file.
*/
@@ -4373,23 +4187,6 @@ NOW: Generate the implementation plan for the task above. Think step by step.`;
return undefined;
}
/**
* Write .mcp.json to a case directory for Claude Code to discover spawn MCP tools.
*/
private writeMcpConfig(casePath: string): void {
const projectRoot = join(__dirname, '..', '..');
const mcpServerPath = join(projectRoot, 'dist', 'mcp-server.js');
const mcpConfig = {
mcpServers: {
'claudeman-spawn': {
command: 'node',
args: [mcpServerPath],
},
},
};
writeFileSync(join(casePath, '.mcp.json'), JSON.stringify(mcpConfig, null, 2) + '\n');
}
private async startScheduledRun(prompt: string, workingDir: string, durationMinutes: number): Promise<ScheduledRun> {
const id = uuidv4();
const now = Date.now();
@@ -5147,10 +4944,6 @@ NOW: Generate the implementation plan for the task above. Think step by step.`;
}
this.respawnControllers.clear();
// Stop spawn orchestrator and all agents
await this.spawnOrchestrator.stopAll();
this.spawnOrchestrator.removeAllListeners();
// Stop all scheduled runs first (they have their own session cleanup)
for (const [id] of this.scheduledRuns) {
await this.stopScheduledRun(id);
-224
View File
@@ -1,224 +0,0 @@
import { describe, it, expect } from 'vitest';
import { buildTaskSpec } from '../src/mcp-server.js';
describe('mcp-server', () => {
describe('buildTaskSpec', () => {
it('should build minimal task spec with required fields only', () => {
const result = buildTaskSpec({
agentId: 'test-001',
name: 'Test Agent',
instructions: 'Do something useful.',
});
expect(result).toContain('---');
expect(result).toContain('agentId: test-001');
expect(result).toContain('name: Test Agent');
expect(result).toContain('Do something useful.');
// Should not contain optional fields
expect(result).not.toContain('type:');
expect(result).not.toContain('priority:');
expect(result).not.toContain('maxTokens:');
expect(result).not.toContain('maxCost:');
expect(result).not.toContain('timeoutMinutes:');
expect(result).not.toContain('contextFiles:');
expect(result).not.toContain('dependsOn:');
});
it('should include all optional fields when provided', () => {
const result = buildTaskSpec({
agentId: 'full-agent',
name: 'Full Agent',
instructions: 'Complete task.',
type: 'implement',
priority: 'high',
maxTokens: 200000,
maxCost: 1.50,
timeoutMinutes: 60,
canModifyParentFiles: true,
contextFiles: ['src/auth.ts', 'src/types.ts'],
dependsOn: ['agent-a', 'agent-b'],
completionPhrase: 'FULL_DONE',
outputFormat: 'structured',
successCriteria: 'All tests pass',
workingDir: '/tmp/workspace',
});
expect(result).toContain('agentId: full-agent');
expect(result).toContain('name: Full Agent');
expect(result).toContain('type: implement');
expect(result).toContain('priority: high');
expect(result).toContain('maxTokens: 200000');
expect(result).toContain('maxCost: 1.5');
expect(result).toContain('timeoutMinutes: 60');
expect(result).toContain('canModifyParentFiles: true');
expect(result).toContain('contextFiles: [src/auth.ts, src/types.ts]');
expect(result).toContain('dependsOn: [agent-a, agent-b]');
expect(result).toContain('completionPhrase: FULL_DONE');
expect(result).toContain('outputFormat: structured');
expect(result).toContain('successCriteria: "All tests pass"');
expect(result).toContain('workingDir: /tmp/workspace');
expect(result).toContain('Complete task.');
});
it('should produce valid YAML frontmatter structure', () => {
const result = buildTaskSpec({
agentId: 'yaml-test',
name: 'YAML Test',
instructions: 'Body content here.',
});
const lines = result.split('\n');
expect(lines[0]).toBe('---');
// Find closing ---
const closingIndex = lines.indexOf('---', 1);
expect(closingIndex).toBeGreaterThan(0);
// Body should come after closing ---
const body = lines.slice(closingIndex + 1).join('\n').trim();
expect(body).toBe('Body content here.');
});
it('should escape double quotes in successCriteria', () => {
const result = buildTaskSpec({
agentId: 'escape-test',
name: 'Escape Test',
instructions: 'test',
successCriteria: 'Output "hello" correctly',
});
expect(result).toContain('successCriteria: "Output \\"hello\\" correctly"');
});
it('should handle empty context files array', () => {
const result = buildTaskSpec({
agentId: 'empty-ctx',
name: 'Empty Context',
instructions: 'test',
contextFiles: [],
});
expect(result).not.toContain('contextFiles:');
});
it('should handle empty dependsOn array', () => {
const result = buildTaskSpec({
agentId: 'empty-deps',
name: 'Empty Deps',
instructions: 'test',
dependsOn: [],
});
expect(result).not.toContain('dependsOn:');
});
it('should handle canModifyParentFiles: false', () => {
const result = buildTaskSpec({
agentId: 'no-modify',
name: 'No Modify',
instructions: 'test',
canModifyParentFiles: false,
});
expect(result).toContain('canModifyParentFiles: false');
});
it('should handle multiline instructions', () => {
const instructions = `# Step 1
Do this first.
# Step 2
Then do this.
## Notes
- Important detail
- Another detail`;
const result = buildTaskSpec({
agentId: 'multiline',
name: 'Multiline',
instructions,
});
expect(result).toContain('# Step 1');
expect(result).toContain('Do this first.');
expect(result).toContain('# Step 2');
expect(result).toContain('- Important detail');
});
it('should be parseable by parseYamlFrontmatter', async () => {
const { parseYamlFrontmatter } = await import('../src/spawn-types.js');
const result = buildTaskSpec({
agentId: 'parse-test',
name: 'Parse Test',
instructions: 'Verify parsing works.',
type: 'explore',
priority: 'high',
maxTokens: 100000,
timeoutMinutes: 15,
});
const parsed = parseYamlFrontmatter(result);
expect(parsed).not.toBeNull();
expect(parsed!.frontmatter.agentId).toBe('parse-test');
expect(parsed!.frontmatter.name).toBe('Parse Test');
expect(parsed!.frontmatter.type).toBe('explore');
expect(parsed!.frontmatter.priority).toBe('high');
expect(parsed!.frontmatter.maxTokens).toBe(100000);
expect(parsed!.frontmatter.timeoutMinutes).toBe(15);
expect(parsed!.body).toBe('Verify parsing works.');
});
it('should produce spec parseable by parseTaskSpecFile', async () => {
const { parseTaskSpecFile } = await import('../src/spawn-types.js');
const result = buildTaskSpec({
agentId: 'full-spec',
name: 'Full Spec Agent',
instructions: 'Do the work.',
type: 'implement',
priority: 'critical',
maxTokens: 250000,
maxCost: 2.0,
timeoutMinutes: 45,
canModifyParentFiles: true,
completionPhrase: 'SPEC_DONE',
outputFormat: 'json',
});
const parsed = parseTaskSpecFile(result, 'fallback-id');
expect(parsed).not.toBeNull();
expect(parsed!.spec.agentId).toBe('full-spec');
expect(parsed!.spec.name).toBe('Full Spec Agent');
expect(parsed!.spec.type).toBe('implement');
expect(parsed!.spec.priority).toBe('critical');
expect(parsed!.spec.maxTokens).toBe(250000);
expect(parsed!.spec.maxCost).toBe(2.0);
expect(parsed!.spec.timeoutMinutes).toBe(45);
expect(parsed!.spec.canModifyParentFiles).toBe(true);
expect(parsed!.spec.completionPhrase).toBe('SPEC_DONE');
expect(parsed!.spec.outputFormat).toBe('json');
expect(parsed!.instructions).toBe('Do the work.');
});
it('should handle contextFiles in parseable format', async () => {
const { parseYamlFrontmatter } = await import('../src/spawn-types.js');
const result = buildTaskSpec({
agentId: 'ctx-parse',
name: 'Context Parse',
instructions: 'test',
contextFiles: ['src/foo.ts', 'src/bar.ts'],
});
const parsed = parseYamlFrontmatter(result);
expect(parsed).not.toBeNull();
expect(Array.isArray(parsed!.frontmatter.contextFiles)).toBe(true);
const files = parsed!.frontmatter.contextFiles as string[];
expect(files).toContain('src/foo.ts');
expect(files).toContain('src/bar.ts');
});
});
});
-968
View File
@@ -1,968 +0,0 @@
import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
import { SpawnOrchestrator, type SessionCreator } from '../src/spawn-orchestrator.js';
import { mkdirSync, writeFileSync, existsSync, rmSync, readFileSync } from 'node:fs';
import { join } from 'node:path';
import { tmpdir } from 'node:os';
/**
* SpawnOrchestrator Tests
*
* Tests the full lifecycle management of spawned agents.
* Uses a temporary directory and mock session creator.
*/
describe('SpawnOrchestrator', () => {
let orchestrator: SpawnOrchestrator;
let testDir: string;
let mockSessionCreator: SessionCreator;
let completionHandlers: Map<string, (phrase: string) => void>;
beforeEach(() => {
testDir = join(tmpdir(), `spawn-test-${Date.now()}-${Math.random().toString(36).slice(2)}`);
mkdirSync(testDir, { recursive: true });
completionHandlers = new Map();
mockSessionCreator = {
createAgentSession: vi.fn().mockResolvedValue({ sessionId: `session-${Date.now()}` }),
writeToSession: vi.fn(),
getSessionTokens: vi.fn().mockReturnValue(0),
getSessionCost: vi.fn().mockReturnValue(0),
stopSession: vi.fn().mockResolvedValue(undefined),
onSessionCompletion: vi.fn().mockImplementation((sessionId, handler) => {
completionHandlers.set(sessionId, handler);
}),
removeSessionCompletionHandler: vi.fn().mockImplementation((sessionId) => {
completionHandlers.delete(sessionId);
}),
};
orchestrator = new SpawnOrchestrator({
casesDir: testDir,
maxConcurrentAgents: 3,
maxSpawnDepth: 2,
defaultTimeoutMinutes: 5,
maxTimeoutMinutes: 10,
progressPollIntervalMs: 60000, // Long interval to avoid interference
});
orchestrator.setSessionCreator(mockSessionCreator);
});
afterEach(() => {
// Stop all agents and clear timers
orchestrator.stopAll().catch(() => {});
orchestrator.removeAllListeners();
// Clean up test directory
if (existsSync(testDir)) {
rmSync(testDir, { recursive: true, force: true });
}
});
function createTaskFile(dir: string, filename: string, content: string): string {
const filePath = join(dir, filename);
mkdirSync(dir, { recursive: true });
writeFileSync(filePath, content);
return filePath;
}
const basicTaskContent = `---
agentId: test-agent-001
name: Test Agent
type: explore
priority: normal
timeoutMinutes: 5
completionPhrase: TEST_DONE
canModifyParentFiles: false
---
# Test Task
Do a simple test.`;
describe('Configuration', () => {
it('should use provided config', () => {
expect(orchestrator.config.maxConcurrentAgents).toBe(3);
expect(orchestrator.config.maxSpawnDepth).toBe(2);
});
it('should update config', () => {
orchestrator.updateConfig({ maxConcurrentAgents: 10 });
expect(orchestrator.config.maxConcurrentAgents).toBe(10);
});
});
describe('handleSpawnRequest', () => {
it('should reject when no session creator is set', async () => {
const noCreator = new SpawnOrchestrator({ casesDir: testDir });
const failHandler = vi.fn();
noCreator.on('failed', failHandler);
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await noCreator.handleSpawnRequest('task.md', 'parent-session', parentDir);
// Should not crash, just log error
expect(failHandler).not.toHaveBeenCalled(); // Silent failure with console.error
});
it('should fail when task file does not exist', async () => {
const failHandler = vi.fn();
orchestrator.on('failed', failHandler);
await orchestrator.handleSpawnRequest('nonexistent.md', 'parent-session', testDir);
expect(failHandler).toHaveBeenCalledWith(
expect.objectContaining({ error: expect.stringContaining('not found') })
);
});
it('should fail when task file cannot be parsed', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'bad.md', 'No frontmatter here');
const failHandler = vi.fn();
orchestrator.on('failed', failHandler);
await orchestrator.handleSpawnRequest('bad.md', 'parent-session', parentDir);
expect(failHandler).toHaveBeenCalledWith(
expect.objectContaining({ error: expect.stringContaining('parse') })
);
});
it('should reject when max depth exceeded', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
const failHandler = vi.fn();
orchestrator.on('failed', failHandler);
// Max depth is 2, so parentDepth=2 means child would be 3 (exceeds)
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir, 2);
expect(failHandler).toHaveBeenCalledWith(
expect.objectContaining({ error: expect.stringContaining('depth') })
);
});
it('should spawn agent and create directory structure', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
// Wait for async initialization
await vi.waitFor(() => {
expect(mockSessionCreator.createAgentSession).toHaveBeenCalled();
});
// Check directory was created
const agentDir = join(testDir, 'spawn-test-agent-001');
expect(existsSync(agentDir)).toBe(true);
expect(existsSync(join(agentDir, 'CLAUDE.md'))).toBe(true);
expect(existsSync(join(agentDir, 'spawn-comms'))).toBe(true);
expect(existsSync(join(agentDir, 'spawn-comms', 'task.md'))).toBe(true);
expect(existsSync(join(agentDir, 'spawn-comms', 'progress.json'))).toBe(true);
expect(existsSync(join(agentDir, 'spawn-comms', 'messages'))).toBe(true);
});
it('should generate proper CLAUDE.md for agent', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(mockSessionCreator.createAgentSession).toHaveBeenCalled();
});
const agentDir = join(testDir, 'spawn-test-agent-001');
const claudeMd = readFileSync(join(agentDir, 'CLAUDE.md'), 'utf-8');
expect(claudeMd).toContain('Agent: Test Agent');
expect(claudeMd).toContain('test-agent-001');
expect(claudeMd).toContain('TEST_DONE');
expect(claudeMd).toContain('# Test Task');
});
it('should emit initializing and started events', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
const initHandler = vi.fn();
const startHandler = vi.fn();
orchestrator.on('initializing', initHandler);
orchestrator.on('started', startHandler);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(startHandler).toHaveBeenCalled();
});
expect(initHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'test-agent-001', name: 'Test Agent' })
);
expect(startHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'test-agent-001', name: 'Test Agent' })
);
});
it('should enforce timeout limits', async () => {
const parentDir = join(testDir, 'parent');
const content = basicTaskContent.replace('timeoutMinutes: 5', 'timeoutMinutes: 999');
createTaskFile(parentDir, 'task.md', content);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
const status = orchestrator.getAgentStatus('test-agent-001');
expect(status).not.toBeNull();
expect(status!.timeoutMinutes).toBe(10); // Capped at maxTimeoutMinutes
});
});
});
describe('Queue Management', () => {
it('should queue agents when concurrency limit reached', async () => {
const parentDir = join(testDir, 'parent');
// Create 4 tasks (limit is 3)
for (let i = 1; i <= 4; i++) {
const content = basicTaskContent
.replace('test-agent-001', `agent-${i}`)
.replace('Test Agent', `Agent ${i}`);
createTaskFile(parentDir, `task${i}.md`, content);
}
const queueHandler = vi.fn();
orchestrator.on('queued', queueHandler);
// Spawn 4 agents
for (let i = 1; i <= 4; i++) {
await orchestrator.handleSpawnRequest(`task${i}.md`, 'parent', parentDir);
}
// Wait for first 3 to start
await vi.waitFor(() => {
expect(mockSessionCreator.createAgentSession).toHaveBeenCalledTimes(3);
});
// 4th should be queued
expect(queueHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'agent-4' })
);
});
it('should order queue by priority', async () => {
const parentDir = join(testDir, 'parent');
// Fill concurrency first
for (let i = 1; i <= 3; i++) {
const content = basicTaskContent
.replace('test-agent-001', `filler-${i}`)
.replace('Test Agent', `Filler ${i}`);
createTaskFile(parentDir, `filler${i}.md`, content);
await orchestrator.handleSpawnRequest(`filler${i}.md`, 'parent', parentDir);
}
await vi.waitFor(() => {
expect(mockSessionCreator.createAgentSession).toHaveBeenCalledTimes(3);
});
// Now add low and high priority
const lowContent = basicTaskContent
.replace('test-agent-001', 'low-agent')
.replace('priority: normal', 'priority: low');
createTaskFile(parentDir, 'low.md', lowContent);
const highContent = basicTaskContent
.replace('test-agent-001', 'high-agent')
.replace('priority: normal', 'priority: critical');
createTaskFile(parentDir, 'high.md', highContent);
await orchestrator.handleSpawnRequest('low.md', 'parent', parentDir);
await orchestrator.handleSpawnRequest('high.md', 'parent', parentDir);
// State should show high priority first in queue
const state = orchestrator.getState();
expect(state.queuedCount).toBe(2);
});
});
describe('cancelAgent', () => {
it('should cancel a running agent', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getAgentStatus('test-agent-001')).not.toBeNull();
});
const cancelHandler = vi.fn();
orchestrator.on('cancelled', cancelHandler);
await orchestrator.cancelAgent('test-agent-001', 'User cancelled');
expect(cancelHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'test-agent-001', reason: 'User cancelled' })
);
});
it('should cancel a queued agent', async () => {
const parentDir = join(testDir, 'parent');
// Fill concurrency
for (let i = 1; i <= 3; i++) {
const content = basicTaskContent
.replace('test-agent-001', `filler-${i}`)
.replace('Test Agent', `Filler ${i}`);
createTaskFile(parentDir, `filler${i}.md`, content);
await orchestrator.handleSpawnRequest(`filler${i}.md`, 'parent', parentDir);
}
// Add one more (queued)
const queuedContent = basicTaskContent.replace('test-agent-001', 'queued-agent');
createTaskFile(parentDir, 'queued.md', queuedContent);
await orchestrator.handleSpawnRequest('queued.md', 'parent', parentDir);
const cancelHandler = vi.fn();
orchestrator.on('cancelled', cancelHandler);
await orchestrator.cancelAgent('queued-agent');
expect(cancelHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'queued-agent' })
);
});
});
describe('sendMessageToAgent', () => {
it('should write message file to comms directory', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getAgentStatus('test-agent-001')).not.toBeNull();
});
await orchestrator.sendMessageToAgent('test-agent-001', 'Focus on JWT');
const messagesDir = join(testDir, 'spawn-test-agent-001', 'spawn-comms', 'messages');
expect(existsSync(join(messagesDir, '001-parent.md'))).toBe(true);
const content = readFileSync(join(messagesDir, '001-parent.md'), 'utf-8');
expect(content).toBe('Focus on JWT');
});
});
describe('getAgentStatus', () => {
it('should return null for unknown agent', () => {
expect(orchestrator.getAgentStatus('nonexistent')).toBeNull();
});
it('should return status for active agent', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
const status = orchestrator.getAgentStatus('test-agent-001');
expect(status).not.toBeNull();
expect(status!.status).toBe('running');
expect(status!.name).toBe('Test Agent');
expect(status!.completionPhrase).toBe('TEST_DONE');
});
});
});
describe('getState', () => {
it('should return complete orchestrator state', () => {
const state = orchestrator.getState();
expect(state.enabled).toBe(true);
expect(state.activeCount).toBe(0);
expect(state.queuedCount).toBe(0);
expect(state.totalSpawned).toBe(0);
expect(state.agents).toEqual([]);
});
it('should update counts after spawn', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
const state = orchestrator.getState();
expect(state.activeCount).toBe(1);
expect(state.totalSpawned).toBe(1);
});
});
});
describe('readAgentMessages', () => {
it('should return empty for unknown agent', () => {
expect(orchestrator.readAgentMessages('nonexistent')).toEqual([]);
});
it('should read messages from comms directory', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getAgentStatus('test-agent-001')).not.toBeNull();
});
// Write a message
await orchestrator.sendMessageToAgent('test-agent-001', 'Hello agent');
const messages = orchestrator.readAgentMessages('test-agent-001');
expect(messages).toHaveLength(1);
expect(messages[0].sender).toBe('parent');
expect(messages[0].content).toBe('Hello agent');
expect(messages[0].sequence).toBe(1);
});
});
describe('triggerSpawn', () => {
it('should spawn from content string', async () => {
const agentId = await orchestrator.triggerSpawn(
basicTaskContent,
'parent-session',
testDir
);
expect(agentId).toBe('test-agent-001');
await vi.waitFor(() => {
expect(mockSessionCreator.createAgentSession).toHaveBeenCalled();
});
});
it('should return null for unparseable content', async () => {
const agentId = await orchestrator.triggerSpawn(
'Not valid YAML frontmatter',
'parent-session',
testDir
);
expect(agentId).toBeNull();
});
});
describe('stopAll', () => {
it('should stop all active agents', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getState().activeCount).toBe(1);
});
await orchestrator.stopAll();
expect(orchestrator.getState().activeCount).toBe(0);
});
});
describe('getPersistedState', () => {
it('should return serializable state', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getState().activeCount).toBe(1);
});
const persisted = orchestrator.getPersistedState();
expect(persisted.config).toBeDefined();
expect(persisted.agents['test-agent-001']).toBeDefined();
expect(persisted.agents['test-agent-001'].completionPhrase).toBe('TEST_DONE');
// Should be JSON-serializable
expect(() => JSON.stringify(persisted)).not.toThrow();
});
});
// ========== Issue Coverage Tests ==========
describe('Cascading Cancellation', () => {
it('should cancel child agents when parent is cancelled', async () => {
const parentDir = join(testDir, 'parent');
// Create parent agent
const parentContent = basicTaskContent
.replace('test-agent-001', 'parent-agent')
.replace('Test Agent', 'Parent Agent');
createTaskFile(parentDir, 'parent.md', parentContent);
await orchestrator.handleSpawnRequest('parent.md', 'user-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getAgentStatus('parent-agent')).not.toBeNull();
});
// Create child agent that depends on parent
const childContent = `---
agentId: child-agent
name: Child Agent
type: explore
priority: normal
timeoutMinutes: 5
completionPhrase: CHILD_DONE
canModifyParentFiles: false
---
# Child Task
Child agent work.`;
createTaskFile(parentDir, 'child.md', childContent);
// Spawn child with parent-agent's session as parent
// Note: In current implementation, we simulate the parent relationship via parentSessionId
await orchestrator.handleSpawnRequest('child.md', 'parent-agent-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getAgentStatus('child-agent')).not.toBeNull();
});
const cancelHandler = vi.fn();
orchestrator.on('cancelled', cancelHandler);
// Cancel parent - this SHOULD also cancel child (if cascading is implemented)
await orchestrator.cancelAgent('parent-agent', 'User cancelled parent');
// Currently, this test documents the EXPECTED behavior.
// The current implementation does NOT cascade cancellations.
// If cascading is implemented, uncomment the assertion below:
// expect(cancelHandler).toHaveBeenCalledTimes(2);
// Current behavior: only parent is cancelled
expect(cancelHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'parent-agent', reason: 'User cancelled parent' })
);
// Verify child is still running (documents current buggy behavior)
const childStatus = orchestrator.getAgentStatus('child-agent');
// When cascading is fixed, this should be 'cancelled' instead of 'running'
expect(childStatus?.status).toBe('running');
});
it('should handle cancellation when no children exist', async () => {
const parentDir = join(testDir, 'parent');
createTaskFile(parentDir, 'task.md', basicTaskContent);
await orchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.waitFor(() => {
expect(orchestrator.getAgentStatus('test-agent-001')).not.toBeNull();
});
const cancelHandler = vi.fn();
orchestrator.on('cancelled', cancelHandler);
// Cancel agent with no children - should work normally
await orchestrator.cancelAgent('test-agent-001', 'Normal cancellation');
expect(cancelHandler).toHaveBeenCalledTimes(1);
expect(cancelHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'test-agent-001', reason: 'Normal cancellation' })
);
// Agent should be cleaned up
expect(orchestrator.getState().activeCount).toBe(0);
});
it('should recursively cancel grandchildren when parent is cancelled', async () => {
const parentDir = join(testDir, 'parent-grandchild');
// Create orchestrator with higher concurrency and depth for this test
const deepOrchestrator = new SpawnOrchestrator({
casesDir: testDir,
maxConcurrentAgents: 5,
maxSpawnDepth: 3,
defaultTimeoutMinutes: 5,
maxTimeoutMinutes: 10,
progressPollIntervalMs: 60000,
});
deepOrchestrator.setSessionCreator(mockSessionCreator);
// Create grandparent agent
const grandparentContent = basicTaskContent
.replace('test-agent-001', 'grandparent-agent')
.replace('Test Agent', 'Grandparent Agent');
createTaskFile(parentDir, 'grandparent.md', grandparentContent);
await deepOrchestrator.handleSpawnRequest('grandparent.md', 'user-session', parentDir);
await vi.waitFor(() => {
expect(deepOrchestrator.getAgentStatus('grandparent-agent')).not.toBeNull();
});
// Create parent agent (child of grandparent)
const parentContent = basicTaskContent
.replace('test-agent-001', 'parent-agent')
.replace('Test Agent', 'Parent Agent');
createTaskFile(parentDir, 'parent.md', parentContent);
await deepOrchestrator.handleSpawnRequest('parent.md', 'grandparent-agent-session', parentDir, 1);
await vi.waitFor(() => {
expect(deepOrchestrator.getAgentStatus('parent-agent')).not.toBeNull();
});
// Create child agent (grandchild of grandparent)
const childContent = basicTaskContent
.replace('test-agent-001', 'child-agent')
.replace('Test Agent', 'Child Agent');
createTaskFile(parentDir, 'child.md', childContent);
await deepOrchestrator.handleSpawnRequest('child.md', 'parent-agent-session', parentDir, 2);
await vi.waitFor(() => {
expect(deepOrchestrator.getAgentStatus('child-agent')).not.toBeNull();
});
// Verify all three agents are running
expect(deepOrchestrator.getState().activeCount).toBe(3);
const cancelHandler = vi.fn();
deepOrchestrator.on('cancelled', cancelHandler);
// Cancel grandparent - this SHOULD cascade to parent and child
await deepOrchestrator.cancelAgent('grandparent-agent', 'User cancelled grandparent');
// Document current behavior: only grandparent is cancelled (no cascade)
expect(cancelHandler).toHaveBeenCalledWith(
expect.objectContaining({ agentId: 'grandparent-agent' })
);
// Current behavior: parent and child are still running (documents the bug)
const parentStatus = deepOrchestrator.getAgentStatus('parent-agent');
const childStatus = deepOrchestrator.getAgentStatus('child-agent');
// When cascading is implemented:
// expect(parentStatus?.status).toBe('cancelled');
// expect(childStatus?.status).toBe('cancelled');
// expect(cancelHandler).toHaveBeenCalledTimes(3);
// Current buggy behavior:
expect(parentStatus?.status).toBe('running');
expect(childStatus?.status).toBe('running');
expect(cancelHandler).toHaveBeenCalledTimes(1);
// Cleanup
await deepOrchestrator.stopAll();
deepOrchestrator.removeAllListeners();
});
});
describe('Resource Budget Validation', () => {
it('should handle negative maxTokens in task spec', async () => {
const parentDir = join(testDir, 'parent');
// Task with negative maxTokens
const content = `---
agentId: negative-tokens-agent
name: Negative Tokens Agent
type: explore
priority: normal
timeoutMinutes: 5
completionPhrase: NEG_DONE
canModifyParentFiles: false
maxTokens: -1000
---
# Test negative tokens`;
createTaskFile(parentDir, 'negative.md', content);
await orchestrator.handleSpawnRequest('negative.md', 'parent-session', parentDir);
await vi.waitFor(() => {
const status = orchestrator.getAgentStatus('negative-tokens-agent');
// Current behavior: negative values are accepted (documents the issue)
// When validation is added, this should either fail or clamp to 0/null
expect(status).not.toBeNull();
});
});
it('should handle zero maxCost in task spec', async () => {
const parentDir = join(testDir, 'parent');
// Task with zero maxCost
const content = `---
agentId: zero-cost-agent
name: Zero Cost Agent
type: explore
priority: normal
timeoutMinutes: 5
completionPhrase: ZERO_DONE
canModifyParentFiles: false
maxCost: 0
---
# Test zero cost`;
createTaskFile(parentDir, 'zero.md', content);
await orchestrator.handleSpawnRequest('zero.md', 'parent-session', parentDir);
await vi.waitFor(() => {
const status = orchestrator.getAgentStatus('zero-cost-agent');
expect(status).not.toBeNull();
// Zero cost budget would immediately trigger 110% threshold check
// on first budget check, causing immediate termination
// This documents potentially problematic behavior
expect(status!.costBudget).toBe(0);
});
});
it('should accept valid budget values', async () => {
const parentDir = join(testDir, 'parent');
const content = `---
agentId: valid-budget-agent
name: Valid Budget Agent
type: explore
priority: normal
timeoutMinutes: 5
completionPhrase: VALID_DONE
canModifyParentFiles: false
maxTokens: 100000
maxCost: 1.50
---
# Test valid budget`;
createTaskFile(parentDir, 'valid.md', content);
await orchestrator.handleSpawnRequest('valid.md', 'parent-session', parentDir);
await vi.waitFor(() => {
const status = orchestrator.getAgentStatus('valid-budget-agent');
expect(status).not.toBeNull();
expect(status!.tokenBudget).toBe(100000);
expect(status!.costBudget).toBe(1.50);
});
});
});
describe('Queue Dependency Handling', () => {
it('should not block independent tasks when one has unmet deps', async () => {
const parentDir = join(testDir, 'parent');
// Fill concurrency first
for (let i = 1; i <= 3; i++) {
const content = basicTaskContent
.replace('test-agent-001', `filler-${i}`)
.replace('Test Agent', `Filler ${i}`);
createTaskFile(parentDir, `filler${i}.md`, content);
await orchestrator.handleSpawnRequest(`filler${i}.md`, 'parent', parentDir);
}
await vi.waitFor(() => {
expect(mockSessionCreator.createAgentSession).toHaveBeenCalledTimes(3);
});
// Add task A that depends on non-existent Task X
const dependentContent = `---
agentId: dependent-agent
name: Dependent Agent
type: explore
priority: normal
timeoutMinutes: 5
completionPhrase: DEP_DONE
canModifyParentFiles: false
dependsOn:
- nonexistent-task-x
---
# Dependent task`;
createTaskFile(parentDir, 'dependent.md', dependentContent);
await orchestrator.handleSpawnRequest('dependent.md', 'parent', parentDir);
// Add Task B with no dependencies
const independentContent = basicTaskContent
.replace('test-agent-001', 'independent-agent')
.replace('Test Agent', 'Independent Agent');
createTaskFile(parentDir, 'independent.md', independentContent);
await orchestrator.handleSpawnRequest('independent.md', 'parent', parentDir);
// Both should be queued
expect(orchestrator.getState().queuedCount).toBe(2);
// Complete one of the filler agents to free up a slot
const completionHandler = completionHandlers.get(
(mockSessionCreator.createAgentSession as ReturnType<typeof vi.fn>).mock.results[0].value.sessionId
);
// Simulate completion by triggering cleanup directly
await orchestrator.cancelAgent('filler-1', 'Test cleanup');
// Wait for queue processing
await vi.waitFor(() => {
// Check if independent-agent started
// Current buggy behavior: dependent-agent blocks the queue
// The test documents this - when fixed, independent-agent should run
const state = orchestrator.getState();
// With the bug: queuedCount stays at 2 or decreases but independent doesn't start
// When fixed: independent-agent should be running
expect(state.activeCount).toBeGreaterThanOrEqual(2);
}, { timeout: 1000 }).catch(() => {
// Expected to fail with current implementation - documents the bug
const state = orchestrator.getState();
// Document current behavior: queue might be stuck
console.log('Queue state (documents starvation bug):', {
activeCount: state.activeCount,
queuedCount: state.queuedCount,
});
});
});
});
describe('Timer Cleanup', () => {
beforeEach(() => {
vi.useFakeTimers();
});
afterEach(() => {
vi.useRealTimers();
});
it('should clear timeout timer on agent completion', async () => {
const parentDir = join(testDir, 'parent-timer1');
mkdirSync(parentDir, { recursive: true });
// Use short timeout for testing
const content = basicTaskContent.replace('timeoutMinutes: 5', 'timeoutMinutes: 1');
createTaskFile(parentDir, 'task.md', content);
// Create a new orchestrator for this test to avoid timer conflicts
const timerOrchestrator = new SpawnOrchestrator({
casesDir: testDir,
maxConcurrentAgents: 3,
maxSpawnDepth: 2,
defaultTimeoutMinutes: 5,
maxTimeoutMinutes: 10,
progressPollIntervalMs: 60000,
});
timerOrchestrator.setSessionCreator(mockSessionCreator);
// Start the spawn request (this sets up timers)
const spawnPromise = timerOrchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
// Run pending timers and promises
await vi.runAllTimersAsync();
await spawnPromise;
// Track that timeout event does NOT fire after cancellation
const timeoutHandler = vi.fn();
timerOrchestrator.on('timeout', timeoutHandler);
// Cancel the agent (which triggers cleanup)
await timerOrchestrator.cancelAgent('test-agent-001', 'Test cleanup');
// Advance timers past the timeout period
await vi.advanceTimersByTimeAsync(2 * 60 * 1000); // 2 minutes
// Timeout should NOT have fired because timer was cleared
expect(timeoutHandler).not.toHaveBeenCalled();
timerOrchestrator.removeAllListeners();
});
it('should clear progress timer on cancellation', async () => {
const parentDir = join(testDir, 'parent-timer2');
mkdirSync(parentDir, { recursive: true });
// Create orchestrator with fast progress polling
const fastPollOrchestrator = new SpawnOrchestrator({
casesDir: testDir,
maxConcurrentAgents: 3,
maxSpawnDepth: 2,
defaultTimeoutMinutes: 5,
maxTimeoutMinutes: 10,
progressPollIntervalMs: 100, // Fast polling
});
fastPollOrchestrator.setSessionCreator(mockSessionCreator);
createTaskFile(parentDir, 'task.md', basicTaskContent);
const spawnPromise = fastPollOrchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.runAllTimersAsync();
await spawnPromise;
const progressHandler = vi.fn();
fastPollOrchestrator.on('progress', progressHandler);
// Cancel the agent
await fastPollOrchestrator.cancelAgent('test-agent-001', 'Test cleanup');
// Clear current call count
progressHandler.mockClear();
// Advance time past several poll intervals
await vi.advanceTimersByTimeAsync(500);
// Progress events should NOT fire after cancellation
expect(progressHandler).not.toHaveBeenCalled();
// Cleanup
fastPollOrchestrator.removeAllListeners();
});
it('should clear warning timer on early completion', async () => {
const parentDir = join(testDir, 'parent-timer3');
mkdirSync(parentDir, { recursive: true });
// Short timeout so warning would fire at ~54 seconds (90% of 1 min)
const content = basicTaskContent.replace('timeoutMinutes: 5', 'timeoutMinutes: 1');
createTaskFile(parentDir, 'task.md', content);
// Create a fresh orchestrator for this test
const warningOrchestrator = new SpawnOrchestrator({
casesDir: testDir,
maxConcurrentAgents: 3,
maxSpawnDepth: 2,
defaultTimeoutMinutes: 5,
maxTimeoutMinutes: 10,
progressPollIntervalMs: 60000,
});
warningOrchestrator.setSessionCreator(mockSessionCreator);
const spawnPromise = warningOrchestrator.handleSpawnRequest('task.md', 'parent-session', parentDir);
await vi.runAllTimersAsync();
await spawnPromise;
// Cancel before warning would fire
await warningOrchestrator.cancelAgent('test-agent-001', 'Early completion');
// Clear the mock
(mockSessionCreator.writeToSession as ReturnType<typeof vi.fn>).mockClear();
// Advance past warning time (54 seconds)
await vi.advanceTimersByTimeAsync(60 * 1000);
// Warning message should NOT have been sent
const writeToSessionCalls = (mockSessionCreator.writeToSession as ReturnType<typeof vi.fn>).mock.calls;
const warningCalls = writeToSessionCalls.filter(
(call: [string, string]) => call[1]?.includes('WARNING') && call[1]?.includes('timeout')
);
expect(warningCalls.length).toBe(0);
warningOrchestrator.removeAllListeners();
});
});
});
-395
View File
@@ -1,395 +0,0 @@
import { describe, it, expect } from 'vitest';
import {
parseYamlFrontmatter,
parseTaskSpecFile,
createDefaultSpawnTaskSpec,
createEmptyAgentProgress,
createInitialSpawnTrackerState,
createDefaultOrchestratorConfig,
serializeSpawnResult,
parseSpawnResult,
AGENT_NAME_MAX_LENGTH,
} from '../src/spawn-types.js';
describe('spawn-types', () => {
describe('parseYamlFrontmatter', () => {
it('should parse basic frontmatter', () => {
const content = `---
name: Test Agent
type: explore
priority: high
---
# Task Body
Do something useful.`;
const result = parseYamlFrontmatter(content);
expect(result).not.toBeNull();
expect(result!.frontmatter.name).toBe('Test Agent');
expect(result!.frontmatter.type).toBe('explore');
expect(result!.frontmatter.priority).toBe('high');
expect(result!.body).toContain('# Task Body');
expect(result!.body).toContain('Do something useful.');
});
it('should parse numbers', () => {
const content = `---
timeoutMinutes: 30
maxCost: 0.50
maxTokens: 150000
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.timeoutMinutes).toBe(30);
expect(result!.frontmatter.maxCost).toBe(0.5);
expect(result!.frontmatter.maxTokens).toBe(150000);
});
it('should parse booleans', () => {
const content = `---
canModifyParentFiles: true
enabled: false
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.canModifyParentFiles).toBe(true);
expect(result!.frontmatter.enabled).toBe(false);
});
it('should parse inline arrays', () => {
const content = `---
dependsOn: [agent-1, agent-2]
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.dependsOn).toEqual(['agent-1', 'agent-2']);
});
it('should parse block arrays', () => {
const content = `---
contextFiles:
- src/auth.ts
- src/middleware.ts
- src/types.ts
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.contextFiles).toEqual(['src/auth.ts', 'src/middleware.ts', 'src/types.ts']);
});
it('should parse empty arrays', () => {
const content = `---
dependsOn: []
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.dependsOn).toEqual([]);
});
it('should parse quoted strings', () => {
const content = `---
name: "Test Agent"
completionPhrase: 'AUTH_DONE'
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.name).toBe('Test Agent');
expect(result!.frontmatter.completionPhrase).toBe('AUTH_DONE');
});
it('should handle null values', () => {
const content = `---
value1: null
value2: ~
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.value1).toBeNull();
expect(result!.frontmatter.value2).toBeNull();
});
it('should handle comments', () => {
const content = `---
# This is a comment
name: Test
# Another comment
type: explore
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.name).toBe('Test');
expect(result!.frontmatter.type).toBe('explore');
});
it('should return null if no frontmatter delimiters', () => {
const content = `No frontmatter here\nJust plain text`;
expect(parseYamlFrontmatter(content)).toBeNull();
});
it('should return null if missing closing delimiter', () => {
const content = `---\nname: Test\nNo closing delimiter`;
expect(parseYamlFrontmatter(content)).toBeNull();
});
it('should handle nested objects', () => {
const content = `---
env:
NODE_ENV: production
DEBUG: true
---
body`;
const result = parseYamlFrontmatter(content);
expect(result!.frontmatter.env).toEqual({ NODE_ENV: 'production', DEBUG: true });
});
});
describe('parseTaskSpecFile', () => {
it('should parse a complete task spec', () => {
const content = `---
agentId: auth-explorer-001
name: Authentication Explorer
type: explore
priority: high
canModifyParentFiles: false
maxTokens: 150000
maxCost: 0.50
timeoutMinutes: 15
resultDelivery: both
completionPhrase: AUTH_EXPLORE_DONE
progressIntervalSeconds: 30
outputFormat: structured
successCriteria: "Document all auth patterns"
---
# Task: Explore Authentication
Analyze the auth system.`;
const result = parseTaskSpecFile(content, 'fallback-id');
expect(result).not.toBeNull();
expect(result!.spec.agentId).toBe('auth-explorer-001');
expect(result!.spec.name).toBe('Authentication Explorer');
expect(result!.spec.type).toBe('explore');
expect(result!.spec.priority).toBe('high');
expect(result!.spec.canModifyParentFiles).toBe(false);
expect(result!.spec.maxTokens).toBe(150000);
expect(result!.spec.maxCost).toBe(0.5);
expect(result!.spec.timeoutMinutes).toBe(15);
expect(result!.spec.completionPhrase).toBe('AUTH_EXPLORE_DONE');
expect(result!.spec.outputFormat).toBe('structured');
expect(result!.instructions).toContain('# Task: Explore Authentication');
});
it('should use defaults for missing fields', () => {
const content = `---
name: Simple Agent
---
Do something.`;
const result = parseTaskSpecFile(content, 'my-fallback');
expect(result).not.toBeNull();
expect(result!.spec.agentId).toBe('my-fallback');
expect(result!.spec.type).toBe('general');
expect(result!.spec.priority).toBe('normal');
expect(result!.spec.timeoutMinutes).toBe(30);
expect(result!.spec.resultDelivery).toBe('both');
expect(result!.spec.outputFormat).toBe('markdown');
expect(result!.spec.canModifyParentFiles).toBe(false);
});
it('should truncate long names', () => {
const longName = 'A'.repeat(100);
const content = `---
name: ${longName}
---
body`;
const result = parseTaskSpecFile(content, 'id');
expect(result!.spec.name.length).toBe(AGENT_NAME_MAX_LENGTH);
});
it('should validate type values', () => {
const content = `---
type: invalid_type
---
body`;
const result = parseTaskSpecFile(content, 'id');
expect(result!.spec.type).toBe('general'); // Falls back to default
});
it('should validate priority values', () => {
const content = `---
priority: super_high
---
body`;
const result = parseTaskSpecFile(content, 'id');
expect(result!.spec.priority).toBe('normal'); // Falls back to default
});
it('should return null for content without frontmatter', () => {
const content = 'No frontmatter at all';
expect(parseTaskSpecFile(content, 'id')).toBeNull();
});
it('should parse contextFiles array', () => {
const content = `---
contextFiles:
- src/auth.ts
- src/types.ts
---
body`;
const result = parseTaskSpecFile(content, 'id');
expect(result!.spec.contextFiles).toEqual(['src/auth.ts', 'src/types.ts']);
});
it('should parse dependsOn array', () => {
const content = `---
dependsOn:
- agent-1
- agent-2
---
body`;
const result = parseTaskSpecFile(content, 'id');
expect(result!.spec.dependsOn).toEqual(['agent-1', 'agent-2']);
});
});
describe('Factory Functions', () => {
it('createDefaultSpawnTaskSpec should generate valid defaults', () => {
const spec = createDefaultSpawnTaskSpec('my-agent');
expect(spec.agentId).toBe('my-agent');
expect(spec.name).toBe('my-agent');
expect(spec.type).toBe('general');
expect(spec.priority).toBe('normal');
expect(spec.timeoutMinutes).toBe(30);
expect(spec.completionPhrase).toContain('MY_AGENT');
expect(spec.completionPhrase).toContain('DONE');
});
it('createDefaultSpawnTaskSpec should sanitize agent ID for completion phrase', () => {
const spec = createDefaultSpawnTaskSpec('my-agent-123');
expect(spec.completionPhrase).toBe('AGENT_MY_AGENT_123_DONE');
});
it('createEmptyAgentProgress should create valid progress', () => {
const progress = createEmptyAgentProgress();
expect(progress.phase).toBe('initializing');
expect(progress.percentComplete).toBe(0);
expect(progress.filesModified).toEqual([]);
expect(progress.tokensUsed).toBe(0);
});
it('createInitialSpawnTrackerState should create valid state', () => {
const state = createInitialSpawnTrackerState();
expect(state.enabled).toBe(false);
expect(state.activeCount).toBe(0);
expect(state.agents).toEqual([]);
});
it('createDefaultOrchestratorConfig should create valid config', () => {
const config = createDefaultOrchestratorConfig();
expect(config.maxConcurrentAgents).toBe(5);
expect(config.maxSpawnDepth).toBe(3);
expect(config.defaultTimeoutMinutes).toBe(30);
expect(config.maxTimeoutMinutes).toBe(120);
expect(config.progressPollIntervalMs).toBe(5000);
});
});
describe('serializeSpawnResult', () => {
it('should serialize a completed result', () => {
const result = {
status: 'completed' as const,
durationMs: 60000,
tokens: { input: 1000, output: 500, total: 1500 },
cost: 0.05,
summary: 'Task completed successfully',
output: '## Result\n\nDetailed output here.',
filesChanged: [
{ path: 'src/auth.ts', action: 'modified' as const, summary: 'Added validation' },
],
agentId: 'test-agent',
completedAt: 1700000000000,
};
const serialized = serializeSpawnResult(result);
expect(serialized).toContain('status: completed');
expect(serialized).toContain('summary: "Task completed successfully"');
expect(serialized).toContain('agentId: test-agent');
expect(serialized).toContain('path: src/auth.ts');
expect(serialized).toContain('## Result');
});
it('should handle empty filesChanged', () => {
const result = {
status: 'failed' as const,
error: 'Something went wrong',
durationMs: 5000,
tokens: { input: 100, output: 50, total: 150 },
cost: 0.01,
summary: 'Failed',
output: 'Error details',
filesChanged: [],
agentId: 'test',
completedAt: Date.now(),
};
const serialized = serializeSpawnResult(result);
expect(serialized).toContain('status: failed');
expect(serialized).toContain('filesChanged: []');
});
});
describe('parseSpawnResult', () => {
it('should parse a result file', () => {
const content = `---
status: completed
summary: "Found 3 auth patterns"
cost: 0.25
---
## Analysis
Detailed findings here.`;
const result = parseSpawnResult(content, 'agent-001', 60000);
expect(result).not.toBeNull();
expect(result!.status).toBe('completed');
expect(result!.summary).toBe('Found 3 auth patterns');
expect(result!.cost).toBe(0.25);
expect(result!.output).toContain('## Analysis');
expect(result!.agentId).toBe('agent-001');
});
it('should handle missing fields with defaults', () => {
const content = `---
status: completed
---
output`;
const result = parseSpawnResult(content, 'agent', 30000);
expect(result!.durationMs).toBe(30000);
expect(result!.cost).toBe(0);
expect(result!.summary).toBe('No summary provided');
expect(result!.filesChanged).toEqual([]);
});
it('should return null for invalid content', () => {
expect(parseSpawnResult('no frontmatter', 'id', 0)).toBeNull();
});
});
});