claudewall
v1.6.0
Published
A personal lessons-learned archive for Claude Code — capture gotchas with /lesson, retrieve them across sessions with /recall, semantic-searched against your prior misses.
Maintainers
Readme
claudewall
A personal lessons-learned archive for Claude Code — capture the gotchas, dead-ends, and corrected assumptions Claude makes during your sessions, and let future-you (in a brand-new session, with no memory of this one) recall the relevant ones via semantic search.
Install
npx claudewall initWhat this does:
- Opens your browser to
claudewall.com/cli/approve - You sign in with GitHub and confirm the device code
- Saves a bearer token to
~/.claudewall/config.json(mode0600) - Installs two slash commands at
~/.claude/commands/:/lesson— capture lessons from the current session/recall— retrieve relevant lessons from past sessions
If you have older claudewall slash commands installed (/wall, /tip, etc.), the installer removes them — they no longer correspond to live endpoints.
Use
Capture — at the end of a session where Claude made a mistake worth remembering:
/lessonClaude scans its own recent assistant turns for specific misses (not generic advice), proposes structured lessons with title / trigger / mistake / correction / tags, and asks which to keep.
Recall — at the start of a new session, before you commit to an approach:
/recall react server actions cachingThe CLI hits the recall endpoint, embeds your query, runs $vectorSearch against your archive, and returns up to 3 lessons above a relevance floor. Claude pulls them into context and works around any prior mistakes.
Re-authorizing
If /api/l/submit or /api/l/recall returns 401, your token has been revoked. Run npx claudewall init again.
Source
github.com/claudewall/source — the cli/ directory in that repo.
License
MIT
