AgentSkillsCN

task-orchestrator

自主代理循环,用于执行、验证并完成各项任务。支持状态转换、子任务管理与审核周期。

SKILL.md
--- frontmatter
name: task-orchestrator
description: Autonomous agent loop for executing, validating, and completing tasks. Handles state transitions, subtask management, and review cycles.

Task Orchestrator

Goal: Drive tasks from todo to completed with strict quality gates.

Core Rules

  1. Validation First: NEVER complete a task without passing build/tests.
  2. Subtask Priority: BLOCKED if subtasks are incomplete.
  3. Status Truth: ALWAYS update status to reflect reality (in-progress when starting).
  4. Context Aware: ALWAYS read linked PRD requirements before execution.

The Loop

1. Claim & Context

Find work and lock it.

bash
# Find next task
npx task-o-matic tasks get-next --status todo

# Set status (replace <ID> with the ID found above)
npx task-o-matic tasks status --id <ID> --status in-progress

# Load Context (Check for prdFile/prdRequirement)
npx task-o-matic tasks show --id <ID>

2. Execute

Perform the work. Use available tools (opencode, edit, bash). If task is complex, split it: npx task-o-matic tasks split --id $TASK_ID

3. Validate

Run project-specific checks.

bash
# Example validation cycle
bun run check-types && bun run build && bun test

4. Review & Fix

Triggers code-reviewer skill if validation passes.

  • If Review Fails: Create fix subtasks.

    bash
    npx task-o-matic tasks create --parent-id $TASK_ID --title "Fix: <issue>" --effort small
    

    Loop back to Step 1 for the fix subtask.

  • If Review Passes: Proceed to completion.

5. Completion

Only when:

  • Code implemented
  • Validation passed
  • Review approved
  • All subtasks completed
bash
npx task-o-matic tasks status --id $TASK_ID --status completed

Helper Scripts

  • scripts/check-status.sh <id>: View task status and subtask tree.