This skill should be used when searching Claude Code session transcripts with semantic understanding. Triggers on queries like "find sessions about X", "when did I work on Y", "search previous conversations". Supports natural language queries with synonym matching.
Search Claude Code session transcripts by combining keyword pre-filtering with semantic evaluation. Finds previous sessions about specific topics, debugging conversations, research tasks, or any past work.
Execute scripts/search.py with the user's query:
python3 scripts/search.py "<query>" [max_results] [max_age_days]
query (required): Natural language search querymax_results (optional, default 10): Maximum results to returnmax_age_days (optional, default 90): How far back to searchThe script performs keyword pre-filtering across all sessions, then extracts meaningful excerpts from top candidates. Output contains a SESSIONS_DATA JSON block.
After receiving the script output, evaluate each session's relevance to the query. Consider:
Assign a relevance score (0-10) to each session based on excerpt content and query intent.
Display the top results (up to max_results) sorted by relevance, formatted as:
### [Relevance: N/10] Project — Date
Summary of what the session was about (1-2 sentences based on excerpts)
`claude --resume <session-id>`
If no relevant results are found, report that and suggest alternative queries.
Sessions are stored as JSONL files in ~/.claude/projects/. Each file contains events with user/assistant messages and tool calls. The search script handles file discovery and text extraction automatically.
To search older sessions or get more results:
/session-search "query" 20 180
(20 results, 180 days lookback)