lessons

LESSONS Command

Lists the feedback memories captured for the current project (and globally) — the "lessons" learned from user corrections.

Context

$ARGUMENTS

Objective

Provide an overview of the self-improvement system: which rules, counter-examples, or preferences have already been recorded to avoid repeating the same mistakes.

Usage

/lessons              # List all feedback memories for the project (read-only)
/lessons <keyword>    # Filter by keyword (e.g., /lessons test, /lessons git)
/lessons --promote    # Capture the lesson just learned into the personal store
/lessons --bootstrap  # One-off: backfill the personal store from existing memories
/lessons --prune      # Keep the personal store under its size budget

The --promote, --bootstrap, and --prune modes feed the personal cross-project lessons store at ~/.claude/rules/lessons.md (loaded into every project by Claude Code). They follow the self-improvement rule: every write is generalized + sanitized + human-confirmed, and the lessons are personal — never committed to any repo.

Modes

--promote (explicit capture — fallback to the reflex)

For when the in-conversation reflex didn't fire but you want to keep a lesson:

1. Take the lesson just learned, generalize it to a one-line principle and sanitize it (strip project/company/person names, paths, URLs, identifiers, verbatim snippets, secrets).
2. Check for recurrence — the store is in your context. If the same principle is already stored, propose bumping its (seen N times) marker instead of adding a duplicate ((seen 2 times) on the first repeat, then N+1).
3. Show it and ask the user to keep / edit / discard (or confirm the bump).
4. On confirmation, write to ~/.claude/rules/lessons.md (create if absent): append the new lesson as a - bullet under its matching ## Topic heading (create the heading if none fits), or apply the recurrence bump. If it can't be generalized, keep it as a local project memory instead.