Harvest insights into agent skills

Query and graduate learning entries from knowledge database

Teams running Claude Code with persistent learning hooks who want to systematically harvest and embed hard-won insights into their agent skills.

• · Search term (for search subcommand)
• · Category filter (implied by subcommand)

• · Learning database status (entry counts, categories)
• · List of all accumulated knowledge
• · Full-text search results ranked by relevance
• · Prescriptive recommendations for agent/skill embedding

• · SQLite database with FTS5 search
• · Python CLI at ~/.claude/scripts/learning-db.py

• · Learning database initialized
• · Entries categorized (design, gotcha, etc.)
• · Target agents/skills exist (for graduation)

• · Graduation assumes prescriptive readiness without manual review
• · FTS5 search may return false positives on common terms
• · Manual learning file parsing breaks if database corrupted

• · Four subcommands with clear separation (status, list, search, graduate)
• · FTS5 full-text search with relevance ranking
• · AI-driven graduation evaluation (prescriptive readiness check)
• · Integration with session-context.py injection hook