retrieval-wasp-drone — Claude Code subagent: what it does
retrieval-wasp-drone — a subagent for an app's retrieval layer: Postgres full-text search, pgvector recall, RRF fusion and recall scoring on a golden query set.
AGPL-3.0-or-later ★ 96 updated 2026-10-18
/plugin marketplace add legioncodeinc/vibe-coding-tools
/plugin install data-and-ai-tools@wasp-nestWhat retrieval-wasp-drone does
The subagent owns how an app finds things. On the default stack that means Neon Postgres: tsvector and websearch_to_tsquery correctness, ranking with ts_rank, semantic recall through pgvector, Reciprocal Rank Fusion to join the two arms, and an optional cross-encoder pass on top. Chunking strategy and recall/precision scoring against a golden query set come with it.
First it works out which implementation the project already runs. A tsvector column and Drizzle schema point at the Postgres guides. A grep-core.ts file points at Deep Lake, where one UNION ALL query spans the memory and sessions tables, <#> handles cosine distance, and BM25/ILIKE quietly takes over once embeddings are off. The skillify branch (src/skillify/*, a KEEP/MERGE/SKIP gate on Haiku) counts as one capability, not the whole domain. Mode routing sits in retrieval-stinger/SKILL.md; audits land under library/requirements/reports/retrieval/.
Scope stops at the query itself. The embedding model and daemon belong to embeddings-runtime-wasp-drone, the vector column and index to vector-store-wasp-drone. Hand it a schema question and it will pass you on.
What it helps with
- diagnose a query that returned nothing useful
- fuse lexical and vector results with RRF
- choose a chunking strategy for retrieval
- score recall and precision on a golden query set
- check whether semantic recall fell back to
ILIKE
Example prompts for the agent
- Tune recall for the
sessionstable and tell me which arms actually ran. - Why did this query miss? Audit the hybrid path in
grep-core.ts. - Add cross-encoder reranking on top of RRF and score it against the golden set.
How to install retrieval-wasp-drone
1Plugin marketplace
/plugin marketplace add legioncodeinc/vibe-coding-tools/plugin install data-and-ai-tools@wasp-nest2Manually: put the files in place
.claude/agents/retrieval-wasp-drone.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/retrieval-wasp-drone.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/retrieval-wasp-drone.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about retrieval-wasp-drone
How do I install retrieval-wasp-drone in Claude Code?
Use the marketplace: run /plugin marketplace add legioncodeinc/vibe-coding-tools, then /plugin install data-and-ai-tools@wasp-nest. The agent file is .claude/agents/retrieval-wasp-drone.md.
What does retrieval-wasp-drone do?
It handles document retrieval in an app: Postgres full-text search, pgvector recall, RRF hybrid fusion, reranking, chunking and quality scoring on golden queries.
How is it different from vector-store-wasp-drone?
The vector column, index and schema are out of scope, as is the embedding model and daemon (embeddings-runtime-wasp-drone). This subagent tunes the query and how the arms are fused.