Planning with Files
A planning skill that writes the state of a long task to disk instead of leaving it in the conversation. It keeps three markdown files: `task_plan.md` for phases and their progress as checkboxes, `findings.md` for research notes and decisions made along the way, `progress.md` for the log of what was done and which tests passed. Hooks feed that plan back on every turn, so you can resume after a `/clear` or a crash without re-explaining where you were. Published by Othman Adi under the MIT licence, installable as a Claude Code plugin or as a skill for other agents.
Strengths
- Survives a /clear or a crash: the plan lives on disk, not in the conversation
- Three files with clear roles (plan, findings, log) rather than a single catch-all document
- MIT licence and content you can read before installing
Limitations
- The plan is fed back on every turn: a growing task_plan.md is paid for in context
- Community repository with a single maintainer: check the hooks before enabling them
Best for
- A rewrite or migration spread over several days, resumed session after session
- A long investigation where decisions made along the way must stay traceable
- Resuming after a /clear or a crash without re-explaining to the agent where you were