← ClaudeAtlas

kb-wiki-readlisted

Find and read wiki pages — fires at session start (reads brain.md HEAD + kb_search) and on any query requiring wiki lookup; evaluates synthesis filing criteria after answering; fallback to full brain.md + index.json walk when kb-search MCP is unavailable.
draakaap/lorewerk · ★ 1 · AI & Automation · score 74
Install: claude install-skill draakaap/lorewerk
# KB Wiki — Read / find pages ## Session start protocol 1. Read `wiki/brain.md` in full — it is a compact dashboard (kept ≤14 KB by wiki-lint Check 11) covering Current work, What changed recently, and the Projects map. 2. Call `kb_search(query, k=5)` with a query derived from the user's first message to find relevant pages. If the first message hasn't arrived yet, defer — wait for the task. For session-start orientation without a specific query, call `kb_top_important(n=10)` instead. Fallback: if `kb_index_status` errors (MCP not connected, db missing), read `brain.md` in full and walk `wiki/index.json` by importance_score — and surface a one-line warning to the user. 3. Load only 1-3 pages — prioritize by RRF score returned by kb_search (not raw importance_score). 4. **Operational logs stay direct-read.** When the task resembles a prior lesson or named pattern, read `wiki/lessons-learned.md` and `wiki/pattern-index.md` directly — they are excluded from the kb-search index. ## Finding a page By semantic search (preferred): ``` kb_search(query=<topic>, k=5) ``` Returns slugs ranked by RRF score. Each result has title, slug, summary, type. Use `kb_get_page(slug)` to load a page. By structured lookup (fallback when kb-search MCP unavailable): ``` Read: wiki/index.json ``` Entries sorted by importance_score. Each has title, slug, summary, type. By keyword search (last resort): ``` Grep pattern: <keyword> Glob filter: wiki/**/*.md Base path: the project root (the directory