← ClaudeAtlas

knowledge-retrievallisted

Cross-store hybrid lexical retrieval, bidirectional [[wikilink]] backlinks, and rename-tracking over the .agent/ knowledge bases. Zero dependency. Use to find relevant knowledge across memory/experiences/references/decisions, discover who links to a note, or safely rewrite links after renaming a note.
Kucell/cortex-agent · ★ 1 · AI & Automation · score 74
Install: claude install-skill Kucell/cortex-agent
# Knowledge Retrieval Skill Zero-dependency knowledge retrieval for the `.agent/` directory, inspired by Obsidian CLI (`backlinks` / `search` primitives) and the Claude-Obsidian plugin's hybrid retrieval — but implemented with pure stdlib (no embeddings on the core path). Complements `graphify` (forward code topology) with markdown semantic links, and generalizes `experience-recall` (experiences-only) to every store. ## When to Use - **Find relevant knowledge** across stores before starting a task → `knowledge-recall` - **Discover impact / who references a note** (backlinks) → `wikilink-index --backlinks` - **Rewrite links after renaming/moving a note** (avoid dead links) → `link-rename` ## Commands ### 0. Unified recall (recommended entry — aggregates the sources below) ```bash node .agent/skills/knowledge-retrieval/scripts/recall.js --query "postcommit hook" [--intent auto|lexical|lesson] [--tags a,b] [--files x] [--limit 8] ``` Single entry point that routes by intent, calls `knowledge-recall` + `experience-recall` in-process, augments ranking with the wikilink backlink signal, then fuses + dedupes + reranks into one result set (each hit tagged with its `source_recallers`). Does **not** replace the sources below — they remain independently usable. `graphify query` is intentionally NOT a result stream here (instruction/LLM-based, not programmatic); its structural role is served by the backlink centrality signal. Output → `metrics/recall-result.json`. Intent routing: