planning-with-files: persistent file-based plans for AI coding agents
planning-with-files is a crash-proof markdown planning layer that keeps AI coding agents’ work state on disk and adds a deterministic completion gate for long-running tasks.
I came across `planning-with-files` from OthmanAdi, and it’s exactly the kind of “agent survival” tooling I want in solo-dev pipelines: persistent, file-backed plans for long-running coding agents.
The core idea is simple: instead of keeping an agent’s plan only in prompt context (which disappears on crashes, timeouts, or reruns), the workflow stores state in markdown on disk. That lets plans survive context loss and supports a `/clear`-style reset without losing the surrounding scaffolding.
Where this becomes useful is when you run an agent through an iterative coding cycle—e.g., “scan repo → update a few modules → run tests → fix failures → repeat”—across multiple invocations. If you’ve ever had an agent lose its place after an interruption, this approach avoids re-deriving intent from scratch.
What to look at first:
- the repo’s description of the persistent markdown plan format (how state is written and updated) - the deterministic completion gate (how the system decides the plan is “done”) - the shared-state mechanism stored on disk for multi-agent coordination - integration/compatibility notes for agent clients (Claude Code, Codex CLI, Cursor, etc.) and the `SKILL.md` standard
Overall, it’s Manus-style in spirit—durable planning artifacts, explicit “done” conditions, and disk-backed shared state—built to work with multiple agent runners (the list mentions 60+ agents). For engineers shipping agent tooling, this is a practical reference for turning “prompt-driven” workflows into restartable systems.
Why it was picked: For a solo AI studio building production Claude Code workflows, “planning-with-files” is directly useful: it introduces crash-proof, persistent file-based planning/state for long-running agent tasks with a deterministic completion gate—exactly the kind of reliability pattern that helps RAG/agent systems ship. It also has strong GitHub signal density via a meaningful relative_trend and clear agent tooling relevance compared with more generic “awesome” lists.