agent-task-queue（Agent 任务队列）

Agent Task Queue

Use this skill when a task needs queue-based orchestration for multiple agents or workers. The bundled runtime is in src/ and covers:

• Priority queue, delayed queue, and dead-letter queue behavior
• Scheduler polling with concurrency limits, retries, and timeouts
• Dependency validation, dependency gating, and result propagation
• Task logs and aggregate metrics
• Pluggable storage through in-memory, SQLite, or Redis backends

Workflow

1. Import TaskQueue and Scheduler from src/index.ts.
2. Pick storage:

- Default InMemoryStorage for local execution or tests - SQLiteStorage for single-node persistence - RedisStorage for distributed workers

1. Register handlers with scheduler.register(taskType, handler).
2. Enqueue tasks with priority, runAt, dependencies, retryPolicy, and timeoutMs as needed.
3. Drive execution via scheduler.tick() or scheduler.start().
4. Inspect queue.logs(), queue.metrics(), queue.getSnapshot(), and queue.get(taskId) for state and traceability.

Key Files

• src/TaskQueue.ts: queue lifecycle, ready/dead-letter snapshots, logs, metrics
• src/Scheduler.ts: polling loop, concurrency control, retries, timeout handling
• src/DependencyManager.ts: DAG validation and dependency result propagation
• src/storage/: storage implementations
• tests/task-queue.test.ts: behavior coverage
• examples/basic.ts: end-to-end usage

Implementation Notes

• Dependency tasks must already exist when a dependent task is enqueued.
• A task becomes runnable only after all dependencies are completed.
• Completed dependency results are stored and exposed to downstream handlers through context.dependencies.
• When retries are exhausted, the task is moved to dead_letter.
• Timeout cancellation uses AbortSignal; long-running handlers should watch context.signal.

Validation

Run:

npm run check

If Redis or SQLite packages are unavailable in the environment, install dependencies first with npm install.