agent-step-sequencer
由 gostlightai 提供
Browser & Automation简介
Multi-step scheduler for in-depth agent requests.
描述
该技能的功能
Multi-step scheduler that detects when a user request requires sequential actions, proposes a step plan, waits for confirmation, then persists and executes each step. Survives gateway resets by writing state to disk and resuming on the next heartbeat. Handles automatic retries and failure recovery without user intervention.
Persisting state to disk and resuming on heartbeat means a mid-execution gateway reset does not lose progress, unlike single-shot agent calls that fail completely if interrupted.
何时使用
- Running a multi-phase research and writing workflow
- Executing sequential API calls with rate-limit delays between steps
- Automating multi-file migrations or setup tasks
- Recovering long-running agent tasks after a connection drop
- Chaining dependent steps where each step's output feeds the next
示例工作流
User asks: Research a topic, write a paper based on the research, then produce an outline of the paper
- 1
Detects 3 sequential actions with output dependencies, proposes a 3-step plan and waits for user confirmation
- 2
User confirms; agent asks whether to apply 2-minute delays between steps for rate-limit safety
- 3
Agent persists state.json with step definitions and immediately invokes step-sequencer-check.py
- 4
Runner executes step-1, verifies requiredOutputs exist, marks DONE, and advances to step-2
- 5
Heartbeat re-invokes check script on schedule; if a step is FAILED, runner retries with a troubleshoot prompt
All steps marked DONE; agent posts a summary with paths to all files created during the workflow
前置要求
- STEP_AGENT_CMD environment variable (command used to invoke the agent, e.g. openclaw agent --message)
安装
clawhub install gostlightai/agent-step-sequencer运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。