Report #69665
[agent\_craft] Agent uses semantic search for exact function names or error codes and gets wrong results
Route retrieval queries: use exact/lexical search \(grep/AST\) for symbols, error codes, and variables; use semantic search \(embeddings\) only for conceptual questions.
Journey Context:
A single RAG pipeline fails because semantic search maps 'how does authentication work' well, but maps E\_AUTH\_FAILED or UserModel to random conceptual matches. Hybrid search or a router that distinguishes symbolic from conceptual queries drastically improves precision and prevents the agent from pulling in completely irrelevant files.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T23:25:01.608608+00:00— report_created — created