okf-knowledge

Files (1)

• SKILL.md

SKILL.md (first lines)

# OKF Knowledge

## Purpose
Make the agent a better coder over time by consulting and growing a persistent
knowledge base of coding lessons (Open Knowledge Format cards), served by the
`okf` MCP server.

## When to use
- Before starting a non-trivial task: `okf_search` for relevant cards
  (patterns, gotchas, prior decisions) and apply what's found.
- After solving something non-obvious — a tricky bug, a Windows quirk, an API
  gotcha, a design decision — capture it with `okf_add` so the next session
  benefits.
- When the user runs `/okf <query>`: search and summarize matching cards; with
  no query, `okf_list` the newest cards.

## Rules
- Search first, add second: run `okf_search` before `okf_add` to avoid
  duplicate cards; extend an existing card with `okf_update` (`append`) when
  the knowledge overlaps.
- Keep cards atomic: one lesson per card, a clear title, correct `type`
  (pattern | snippet | gotcha | decision | howto | reference), and 2-5 tags.
- Include a minimal code example in the body when the lesson is about code.
- Record the `source` (URL or file path) when the knowledge came from
  somewhere specific.
- Link related cards via `links` so knowledge forms a graph, not a pile.
- Never store secrets (API keys, tokens, passwords) in cards.

From tattooinmtl/NimAgent2, MIT.

More from tattooinmtl/NimAgent2