peterfile/devpilot-agents
Overview
This skill runs a Python-based sequential orchestrator that executes Kiro spec tasks one at a time with tmux visibility. It automates a full orchestration loop, handling assignment, dispatch, synchronous execution, state updates, and final reporting without pausing for input. Use it when you need safe, throttled, serial task execution to avoid rate limits or quota exhaustion.
How this skill works
The orchestrator invokes scripts/sequential_loop.py via the shell and loops until all tasks are complete or a HALT signal appears. Each iteration finds the next incomplete task, ensures an LLM-determined assignment exists, dispatches the task into a tmux window, waits for synchronous completion, updates SEQUENTIAL_STATE.json and progress logs, then sleeps before the next iteration. The entire flow runs in a single blocking command and exits with a final summary and clear exit code.
When to use it
- You must avoid parallel execution to prevent quota exhaustion or API rate limits.
- Running long overnight jobs that must proceed serially and be observable in tmux.
- A task set is throttled and must run with delay between iterations.
- You want deterministic, auditable updates to a SEQUENTIAL_STATE.json progress file.
- A user explicitly requests: 'Run sequential orchestration for <spec>'.
Best practices
- Run the orchestrator with a single blocking shell command and set a 2-hour timeout for long runs.
- Provide --delay and --max-iterations to control throttling and safety limits.
- Keep CODEAGENT_NO_TMUX=1 if tmux is unavailable or you prefer plain stdout.
- Monitor .kiro/specs/SEQUENTIAL_STATE.json and SEQUENTIAL_PROGRESS.md for live status.
- Handle HALT signals promptly; HALT stops the loop and requires human intervention.
Example use cases
- Throttled API migration where each subtask must finish before the next begins to stay under rate limits.
- Overnight test-and-deploy pipeline that needs serial, visible steps to debug intermittent failures.
- Running a prioritized spec where tasks must be assigned by an LLM sub-agent and executed in order.
- Small-team workflows that require a single tmux session for live monitoring and easy attachment.
- Recovering from transient failures by re-running the orchestrator to resume from saved state.
FAQ
Set the environment variable CODEAGENT_NO_TMUX=1 before running the command to disable tmux windows.
What exit codes indicate termination reasons?
Exit 0 means all tasks completed, 1 means max iterations reached, and 2 means the run halted waiting for human input.
2 skills
This skill orchestrates tasks sequentially using a Python-based runner with tmux visibility to ensure orderly, throttled execution.
This skill helps you generate a complete Kiro spec pack (.kiro/specs) including requirements with EARS, design, and executable tasks.
