.pi/skills/planning-with-files/README.md
Work like Manus - Use persistent markdown files as your "working memory on disk."
A Pi Coding Agent package that ships both:
pi install npm:@tomxprime/planning-with-files
# From the planning-with-files repo root
pi install ./.pi/skills/planning-with-files
Or add to .pi/settings.json:
{
"packages": ["./path/to/planning-with-files/.pi/skills/planning-with-files"]
}
Pi discovers the skill and extension from the installed package.
Start with:
Use the planning-with-files skill to help me with this task.
Or:
/skill:planning-with-files
The bundled extension maps Claude-style behavior onto Pi events:
session_start - session catchupbefore_agent_start - plan reminder/injectiontool_call - pre-tool recitation equivalenttool_result - post-write reminderagent_end - incomplete-task auto-continue (limit 3)session_before_compact - pre-compaction reminderAttestation is supported. If task_plan.md differs from approved hash, plan injection is blocked with:
[planning-with-files] [PLAN TAMPERED - injection blocked]
planningWithFiles.mode supports:
auto (default): DeepSeek -> cache-safe, others -> parityparity: full dynamic hook-equivalent behaviorcache-safe: fixed reminder strings for KV-cache stabilitynotify: notification-only modeConfigure via env:
PWF_MODE=cache-safe pi
Or settings:
{
"planningWithFiles": {
"mode": "auto"
}
}
/plan-status/plan-attest [--show|--clear]/plan-goal <text|default|clear>/plan-loop [interval] [prompt] (stop to cancel)If needed, run catchup manually:
python3 .pi/skills/planning-with-files/scripts/session-catchup.py .
The skill workflow still centers on three files in your project:
your-project/
├── task_plan.md
├── findings.md
└── progress.md