knowledge-manager
Manages a structured knowledge base of patterns, troubleshooting guides, best practices, and workflo
提供方 fubira|开源
Knowledge Manager Skill
Manage a structured knowledge base. Progressive disclosure: reference knowledge only when relevant, keeping CLAUDE.md minimal.
Activation Triggers
- Discovered a reusable solution (automatic)
- Solved a non-obvious technical problem (automatic)
- Established a new coding convention or pattern (automatic)
Knowledge Base Location
All entries live under RESOURCES/AI_KNOWLEDGE/ in the Obsidian vault (vault root path is defined in the global CLAUDE.md "Obsidian" section; referred to as AI_KNOWLEDGE/ below). Global, cross-project — not project-local, and shared with Codex, so keep entries tool-neutral.
Categories
| Category | Path | When to Record |
|---|---|---|
| Patterns | AI_KNOWLEDGE/patterns/ | Reusable design patterns, architecture |
| Troubleshooting | AI_KNOWLEDGE/troubleshooting/ | Non-obvious technical problem solutions |
| Best Practices | AI_KNOWLEDGE/best-practices/ | Coding standards, quality guidelines |
| Workflows | AI_KNOWLEDGE/workflows/ | Dev processes, CI/CD, operational procedures |
Recording Process
- Detect: Identify valuable insights during development
- Evaluate: Assess on 3 axes — Reusability, Impact, Learning Value (record if 2/3 are Medium+)
- Record: Check category INDEX.md → deduplicate → create entry with front matter per
AI_KNOWLEDGE/FRONTMATTER.md→ update that category's INDEX.md - User Approval: Present summary, category, evaluation, and usage examples; create only after approval
Entry Structure
Keep entries concise (< 100 lines). Common elements:
- Overview: 1-2 sentence summary
- Context/Rationale: When/why this applies
- Details: Commands, code, or steps (language-tagged fenced blocks)
- Pitfalls: Known failure modes
- References: External docs or related entries
Infer exact structure from the category — no fixed template required. See existing entries in AI_KNOWLEDGE/ for style.
Search and Retrieval
- Check relevant category INDEX.md
- Read only the specific files needed — never read a category in bulk
- Provide answers with source references
Maintenance
- Periodically update INDEX files, archive low-value entries, consolidate duplicates