deep-search

name	deep-search
description	Multi-strategy parallel search across the codebase. Combines symbol search, text patterns, file structure, imports, and git history for comprehensive results. Activate when #ds appears anywhere in the user message.
argument-hint	[query]
disable-model-invocation	false
allowed-tools	Read, Glob, Grep, Bash, Task

Search the codebase for: $ARGUMENTS

Execute a multi-strategy parallel search to find code comprehensively.

Step 1 — Analyze the query:

• What did the user literally ask?
• What are they actually trying to accomplish?
• What result would let them proceed immediately?

Step 2 — Launch parallel searches (at least 3 strategies simultaneously):

1. Symbol search — Grep for function names, class names, type definitions, variable declarations.
2. Text search — Grep for string literals, error messages, log output, comments.
3. File pattern search — Glob for file naming conventions and directory structure.
4. Import search — Grep for import/require/use statements to trace dependency chains.
5. Git history search — If git is available, search git log for when code was added, modified, or deleted.

Use deepsearch agent for parallel execution when the query is broad. Use direct Grep and Glob when the query is specific.

Step 3 — Synthesize results:

• Deduplicate findings across strategies.
• Rank by relevance: exact matches first, then partial, then contextual.
• For each result, explain WHY it is relevant to the query.
• Rate overall confidence: HIGH (exact match), MEDIUM (likely match), LOW (best guess).

Step 4 — Provide actionable output:

• If the user asked "where is X": provide the exact file and line.
• If the user asked "how does X work": trace the flow and explain it.
• If the user asked "find all X": provide a comprehensive list.

End with:

• Files (absolute path and relevance for each)
• Answer (direct answer to the actual need)
• Confidence
• Alternative queries (if confidence is not HIGH)