Keeps your agent's plan, findings and progress in markdown files on disk, so long tasks survive /clear, crashes and compaction.
by OthmanAdi· AI & agentsUnclaimed · Yours? Claim it
4 images
About
Planning with Files is a skill and Claude Code plugin by Ahmad Othman Ammar Adi (OthmanAdi) that gives your agent a working memory on disk. It follows the file-based planning pattern Manus described: before a multi-step task, the agent writes task_plan.md, findings.md and progress.md in your project and keeps them up to date as it works. Lifecycle hooks put the current plan back in front of the agent every turn, so after /clear, a crash or compaction it picks up at the current phase. What it teaches the agent - Create the plan first: phases with checkboxes in task_plan.md before any work starts - Save research notes and decisions to findings.md after every two view or browser operations - Log progress, test results and every error in progress.md, and never repeat a failed attempt - Re-read the plan before big decisions, and check off each phase when it's done - Recover after a context reset by reading the planning files back from disk What the Claude Code plugin adds - Lifecycle hooks (SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, PreCompact and Stop) that re-inject the active plan - Slash commands: /plan to start, a status report on the current phase, /plan-doctor to check the hooks really fire, /plan-attest to lock the plan with a SHA-256 hash, and /plan-goal and /plan-loop to keep going until the plan is complete - Autonomous and gated modes for long runs; gated mode holds the agent's stop while a phase is still in progress, with a cap so it never traps a session - Separate plans for parallel tasks under .planning/, each pinned with PLAN_ID, and per-agent run ledgers for multi-agent work - Translations in Arabic, German, Spanish, Simplified Chinese and Traditional Chinese (/plan-ar, /plan-de, /plan-es, /plan-zh, /plan-zht) How to install it In Claude Code, add the maker's marketplace with /plugin marketplace add OthmanAdi/planning-with-files, then run /plugin install planning-with-files@planning-with-files. That route ships the skill, its hooks and its slash commands. For other agents, npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g installs the skill, and the README has steps for Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Hermes Agent and more. Good to know - No sign-in: it runs inside your agent and writes plain markdown files to your project, gitignored by default. - It has no network upload path. Its catch-up modes read your agent's local session records for the same project only when you ask. - Skill-only installs can end up without hooks; the plugin route ships them, and /plan-doctor checks they fire. - It's MIT licensed.
Once it’s added
“Plan this task before you start: break it into phases in task_plan.md.”
“We just cleared the context. Read the plan and carry on from the current phase.”
“Start a gated plan that doesn't stop until every phase is complete.”
“Lock the plan so any change to it gets flagged.”
“Check that the planning hooks are firing.”
“Set up a separate plan for this parallel task.”
Before you add it
Reads only
Your planning files, at the start of each turn · Same-project local session records, only when you ask for a catch-up
Makes changes · your agent can ask first
task_plan.md, findings.md and progress.md in your project · A SHA-256 attestation when you lock the plan · Per-agent run ledgers under .planning/ for multi-agent runs
Each with its own page
From people who use it
Step by step
No reviews yet. Used Planning with Files? Say how it went.
Write a reviewRuns on your machine.
Runs on your machine.
/plugin marketplace add OthmanAdi/planning-with-files /plugin install planning-with-files@planning-with-files
Scanned and passed · checked 45 min ago
Nobody has shared how they use Planning with Files yet.
Plugins people install alongside Planning with Files.
Nothing new reaches your agent until a person has looked at it.
Scanned on every change, and again when our rules improve.
Claude CodeSupported
OpenClawSupportedTell us and a person on the trust team will look at it. Or email [email protected].
Plugins people install alongside Planning with Files.
Two debugging subagents and three commands that find the root cause of errors, set up error tracking and run multi-agent code reviews.
ByteDance's local MCP server that lets your agent run any shell command or script on your machine, with your own permissions.
A self-learning skill layer for Claude Code: distills skills from your real sessions, updates them as you work and prunes unused ones.
Security scan
Passed
Safe to install in every agent it supports. Claude plugins.
