research-search

Search Research Memory

Search .ai-memory/research.jsonl for existing findings before starting new research.

Steps

1. Parse query: $ARGUMENTS contains the search keywords. If empty, ask the user what they're looking for.

2. Read .ai-memory/research.jsonl. If it doesn't exist or is empty, tell the user: "No research findings recorded yet. Proceed with your investigation."

3. Semantic evaluation — YOU are the search engine. Read ALL entries and evaluate semantic relevance to the query:

   • Match on: synonyms ("DB" = "database"), abbreviations ("JWT" = "JSON Web Token"), conceptual overlap ("error handling" ≈ "exception strategy"), and architectural relevance.
   • Do NOT rely on exact keyword matching — use your understanding of meaning.

4. Show the most relevant results (up to 5) with full finding details and staleness assessment:

   For each match:

   ### [<STALENESS_BADGE>] <topic>
   **Tags**: <tags>  |  **Confidence**: <confidence>  |  **Date**: <date>
   **Finding**: <full finding text>
   **Source**: <source_tool> — <source_context>
   

   Staleness badges:

   • [FRESH] — staleness is "stable", or staleness is "versioned" and version_anchored matches a recent/current version
   • [CHECK VERSION] — staleness is "versioned" (user should verify version_anchored matches their current version)
   • [VERIFY] — staleness is "volatile" (user should verify finding is still accurate)

5. Recommendation:

   • If a [FRESH] match with confidence >= 0.8 is found: "Recommendation: Use this finding directly."
   • If a [CHECK VERSION] match is found: "Recommendation: Verify your current version matches <version_anchored> before using."
   • If a [VERIFY] match is found: "Recommendation: Use as a starting hypothesis but verify — this finding may have changed."
   • If no matches: "No matching research found. Proceed with your investigation, then save findings with /project-memory:research-save."