long-running-agent

To convert a project specification into a phased workflow, this skill breaks down planning documents into logical phases with granular, testable tasks and verification checkpoints. It automatically generates compliant task JSON files and customized agent prompts for autonomous execution.

A long-running agent workflow is a structured, phase-based task sequence for multi-session software development requiring mandatory verification checkpoints. You need this process when executing complex project buildouts autonomously to ensure strict progress tracking and prevent skipped requirements.

You can set up an autonomous agent prompt by providing your project spec to this skill, which then generates a ready-to-use claude-prompt.md file. This prompt enforces sequential phase progression, per-task git checkpointing, and progress validation aligned with Anthropic's long-running agent engineering patterns.

Yes, you can use existing planning documents directly for task decomposition. The skill processes your attached project SPEC.md automatically, splitting it into implementation phases with granular tasks and verification steps without requiring you to manually restructure the document beforehand.

The best way to track progress and enforce git checkpointing is by generating a task JSON file and customized agent prompt that mandate per-task git commits. This approach aligns with long-running agent best practices to validate progress sequentially across each implementation phase.