AI Skill MartBrowse skillsPromptsProgramsLearnWhat is a Skill?
 
← Back to list

Planning with Files

Keeps task_plan.md, findings.md and progress.md on disk so an agent's multi-step work survives /clear, compaction and context loss.

AutomationIntermediate★ 26,415⑂ 2,209AI score 10/10Last updated: Aug 29, 2026

What it does

Brings the Manus-style "working memory on disk" pattern to Claude Code. Before any complex task it creates task_plan.md (phases and status), findings.md (research) and progress.md (session log) in your project root, then enforces updates after every phase.

  • Automatic context injection: UserPromptSubmit, PreToolUse and PreCompact hooks re-inject the plan head so the agent stays on goal.
  • Session recovery: after /clear, session-catchup.py finds unsynced context and reconciles it with git diff.
  • 3-strike error protocol: never repeat a failing action, escalate to the user after three attempts.
  • Parallel plans: isolated .planning/<date-slug>/ directories, pinned via PLAN_ID or PWF_PLAN_ROOT.
  • Safety layers: plan content is wrapped in BEGIN/END data delimiters, and SHA-256 attestation blocks injection if the plan file is modified outside the approved flow.

Who it's for

  • Anyone running long agent sessions (refactors, migrations, builds) with dozens of tool calls
  • Users frustrated by agents losing the goal after auto-compaction
  • Researchers who want findings written to files, not buried in chat scrollback
  • Teams juggling several concurrent work threads in one repository

Examples

  1. Large refactor: init-session.sh "Backend Refactor" creates a six-phase plan; even after /clear, the next turn re-reads the plan and picks up where it left off.
  2. Long research task: the 2-action rule forces findings to findings.md after every two view/search operations, so screenshot and PDF content becomes durable text.
  3. Unattended run: init-session.sh --gated "Build Pipeline" enables the completion gate so the agent won't stop while a phase is still in_progress, with a block cap and stall detection as guardrails.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/OthmanAdi/planning-with-files/HEAD/.pi/skills/planning-with-files/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on AI Skill Mart.
Copy the .pi/skills/planning-with-files folder from the GitHub repo OthmanAdi/planning-with-files into my ~/.claude/skills/pi-planning-with-files/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/OthmanAdi/planning-with-files.git /tmp/planning-with-files && mkdir -p ~/.claude/skills && cp -r /tmp/planning-with-files/.pi/skills/planning-with-files ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal and clone the repo: git clone https://github.com/OthmanAdi/planning-with-files.git /tmp/planning-with-files
  2. Copy the skill into place: mkdir -p ~/.claude/skills && cp -r /tmp/planning-with-files/.pi/skills/planning-with-files ~/.claude/skills/
  3. Make the helper scripts executable: chmod +x ~/.claude/skills/planning-with-files/scripts/*.sh
  4. If you also want the /plan-goal and /plan-loop slash commands, use the plugin route instead: run /plugin marketplace add OthmanAdi/planning-with-files in Claude Code, then /plugin install.
  5. Restart Claude Code, open any project and ask it to "plan out this task"; confirm that task_plan.md appears in the project root.
  6. If the hooks seem silent or behave oddly, run the self-check: sh ~/.claude/skills/planning-with-files/scripts/plan-doctor.sh
  7. (Optional) Install python3 if you want the post-/clear session recovery feature.
View source on GitHub ↗License: MIT