# Prowl Durable Knowledge

> Prowl durable knowledge

- Skill: `neur0map/prowl-durable-knowledge` (Agent Skill)
- Install (CLI): `npx skillmds@latest add neur0map/prowl-durable-knowledge`
- Raw SKILL.md: https://api.skillmd.com/api/skills/neur0map/prowl-durable-knowledge/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: neur0map (https://skillmd.com/u/neur0map)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/neur0map/prowl-durable-knowledge

---


# Prowl durable knowledge

An agent's context window is thrown away at the end of a session, so whatever you
figured out about the project evaporates unless you write it down where the next
agent will find it. prowl-agent stores accepted decisions, concepts, and gotchas
as portable OKF Markdown, separate from the disposable index, and serves them
back through `search`/`context` so understanding compounds across agents instead
of being rediscovered every time.

## When to capture

Capture a note when you resolve something that cost real effort and will matter
again: why a design is the way it is, a convention that is not obvious from the
code, a trap that bit you, or the answer to a question the user or a teammate
will ask again.

## Workflow

1. First run only: `prowl-agent knowledge init`.
2. Write the note as OKF Markdown (a decision, concept, claim, or playbook) and
   propose it: `prowl-agent knowledge propose --file note.md --target decisions/<slug>.md`.
   The command prints a deterministic diff before anything is stored.
3. Accept it once reviewed: `prowl-agent knowledge accept <proposal-id>`.
4. Check health anytime: `prowl-agent knowledge list` and `prowl-agent knowledge lint`.

Anchor claims to real code where you can; prowl flags a note as stale when the
code region it points at changes, so future readers know when to re-verify. Keep
notes short and evidence-backed; do not capture routine steps or anything the
code already states plainly.