@skillbolt/execute

Execution engine for Skillbolt. Runs tasks via AI agent SDKs with isolated sessions, DAG orchestration, and log streaming.

Features

• RunContext — Isolated run directories with skill copying, file staging, meta/result tracking
• SkillOrchestrator — Phase-based parallel execution with plan generation and visualization
• AgentClient abstraction — Pluggable agent backends (Mock, Claude, custom)
• Prompt builders — Isolated and direct executor prompt generation
• Summary extraction — Parse <execution_summary> tags from agent responses
• Run management — List, get, and clean up old runs

Quick Start

import { SkillOrchestrator, RunContext, MockAgentClient } from '@skillbolt/execute';

const runContext = RunContext.create('fix the login bug', { mode: 'dag' });

const orchestrator = new SkillOrchestrator({
  maxConcurrent: 6,
  nodeTimeout: 600,
  runContext,
});

const result = await orchestrator.runWithVisualizer({
  task: 'fix the login bug',
  skillNames: ['debugger', 'test-writer'],
  visualizer, // implements VisualizerProtocol
});

console.log(result.status); // 'completed' | 'partial' | 'failed'

CLI

skill run "fix the login bug" --skills debugger,test-writer --mode dag
skill run "migrate database" --plan-only
skill run "generate report" --mode freestyle

Execution Modes

| Mode | Description | | ----------- | -------------------------------------------------- | | dag | Auto-generate DAG plan, execute in parallel phases | | freestyle | Agent chooses which skills to use | | direct | Single skill execution | | baseline | No skills, agent works directly |

API

| Export | Description | | ----------------------------- | ---------------------------------------- | | SkillOrchestrator | Main orchestrator with plan→execute flow | | RunContext | Isolated run directory management | | RunManager | List, get, cleanup runs | | MockAgentClient | Testing client with canned responses | | ClaudeAgentClient | Claude Agent SDK client (stub) | | buildIsolatedExecutorPrompt | Build prompt for skill execution | | buildDirectExecutorPrompt | Build prompt for direct execution | | extractExecutionSummary | Parse execution results |

License

MIT