claudegoodies
Skill

wiki-retrieve

From AgriciDaniel

Hybrid retrieval primitive for the Compound Vault. Replaces the v1.6 static hot→index→drill read order with contextual-prefix + BM25 + cosine-rerank, modeled on Anthropic's Sept 2024 Contextual Retrieval research (35-49-67% retrieval-failure reduction). Opt-in via `bash bin/setup-retrieve.sh`; feature-detected by wiki-query and autoresearch. Triggers on: retrieve, hybrid retrieval, BM25, rerank, contextual retrieval, search the chunks, chunk search, vault search, semantic search, what chunks match, find relevant passages.

Adds chunk-level hybrid retrieval (contextual prefix + BM25 + cosine rerank) over a wiki vault, replacing page-level reads.

Use it when

  • Answers live in specific passages rather than whole wiki pages
  • Building or extending wiki-query / autoresearch skills that need chunk search
  • Want lower retrieval-failure rate than the v1.6 hot→index→drill flow
  • Already run other Compound Vault skills and want opt-in upgrade

Skip it if

  • Requires manual opt-in setup script; does nothing until provisioned
  • Non-synthetic prefix tiers send wiki page bodies to Anthropic API or claude CLI
  • Dense rerank depends on a local ollama server running nomic-embed-text
  • Tied specifically to the Compound Vault / claude-obsidian wiki structure

Facts

Repository
AgriciDaniel/claude-obsidian
Status
Actively maintained
Last commit
Declared tools
Read Bash
Source file
skills/wiki-retrieve/SKILL.md

Source preview

The instructions Claude Code reads when this skill runs.

# wiki-retrieve: Hybrid Retrieval over the Vault

The v1.6 query path was `Read(hot.md) → Read(index.md) → Read(3-5 pages) → synthesize`. It worked, but page-level granularity loses to chunk-level granularity any time the answer lives in a specific passage rather than a whole page. The v1.7 `wiki-retrieve` skill is the chunk-level upgrade — opt-in, feature-gated, and replaces nothing if you don't run the setup.

**Origin**: This skill is original to claude-obsidian. There is no upstream kepano equivalent. The technique is from [Anthropic's Sept 2024 Contextual Retrieval research](https://www.anthropic.com/news/contextual-retrieval) — we implement it as agent-skill plumbing.

---

## Data privacy (v1.7.1+)

Tier 1 (Anthropic API) and tier 2 (claude CLI subprocess) of the contextual-prefix generator send **wiki page bodies off-machine**. As of v1.7.1, both tiers are GATED behind explicit user consent at two layers:

- `scripts/contextual-prefix.py --allow-egress` (default off). Without the flag, `pick_prefix_tier()` returns `"synthetic"` regardless of `ANTHROPIC_API_KEY` or `claude` binary presence.
- `bin/setup-retrieve.sh` prompts before any non-synthetic Stage 1 run; default is abort.

To run fully on-machine (tier 3 synthetic prefix + local ollama rerank), use `bash bin/setup-retrieve.sh --no-llm`. This is also the effective behavior if you decline the consent prompt or omit
View full source on GitHub →

Other skills