Agent Skills: Intelligent Code Search

Intelligent Code Search

Use the cheapest tool that answers the question with verified file references. For code structure, that means ast-grep before text search. For semantic flow, that means targeted entry points first, then WarpGrep or another semantic search. Do not read the whole repo because the user sounds curious. Curiosity is not a query plan.

Search Tool Order

1. ast-grep / sg — code structure: language constructs, call shapes, functions containing or missing code, AST-aware refactor candidates.
2. rg — exact text: strings, comments, docs, logs, config keys, unknown identifiers where structure is irrelevant.
3. fd — file and directory discovery.
4. WarpGrep / semantic search — multi-hop code flow and architecture questions after entry points are known.
5. grep / find — only when modern tools are unavailable.

Critical Workflow Rules

• Classify the query first: structural code search, text search, file discovery, or semantic flow. Do not blindly start with rg.
• For structural code searches, test ast-grep on a small pattern or inline rule, then run it on the scoped tree.
• For trace-flow and architecture questions, read domain docs first when present: CONTEXT.md, CONTEXT-MAP.md, nearest */CONTEXT.md, and relevant docs/adr/*.md files.
• Do not read the whole repo indiscriminately. Convert vague asks into a scoped map question or ask for scope.
• Trace callers, callees, shared types/messages, and data/control flow across files. Follow only enough files or line ranges to verify the map.
• Separate known facts from guesses. List unknowns explicitly instead of filling gaps.
• For vague requests like "read this repo and explain everything", refuse the full dump, offer a zoom-out map, and say the final summary will separate verified facts from guesses/unknowns.

ast-grep Quick Use

Check the executable. Prefer ast-grep; use sg if that is what the system has.

command -v ast-grep || command -v sg

Pattern search:

ast-grep run --pattern 'console.log($$$)' --lang javascript .
ast-grep run --pattern 'func $NAME($$$) { $$$ }' --lang go .
ast-grep run --pattern 'def $NAME($$$): $$$' --lang python . --json

Relational or composite rule search:

ast-grep scan --inline-rules 'id: async-without-try
language: javascript
rule:
  all:
    - kind: function_declaration
    - has:
        pattern: await $EXPR
        stopBy: end
    - not:
        has:
          pattern: try { $$$ } catch ($E) { $$$ }
          stopBy: end
' .

Debug pattern parsing:

ast-grep run --pattern 'class $NAME { $$$BODY }' --lang javascript --debug-query=pattern
ast-grep run --pattern 'class User { constructor() {} }' --lang javascript --debug-query=cst

Rule authoring details live in references/ast-grep-rule-reference.md.

When to Use Which Tool

ast-grep:

• "Find all async functions without try/catch."
• "Find calls to foo($A, $B) inside class methods."
• "Find React components that call useEffect with an empty dependency list."
• "Find handlers that return errors without wrapping."
• Structural rewrite candidate mapping.

rg:

• "Find the string NEXT_PUBLIC_API_URL."
• "Search docs for OpenAI."
• "Find TODO comments."
• "Where is UserService mentioned?" when node shape does not matter.

WarpGrep / semantic search:

• "How does auth flow work?"
• "Trace data from API to DB."
• "Find all places where permissions are checked" when naming is inconsistent.
• Large repos where the answer requires multi-hop reasoning.

Smart Explore after narrowing:

• "What functions are in this file?"
• "Show me this function's source."
• "Extract these 3 symbols from the files the code map found."

Query Formulation

Good semantic-flow queries

How does authentication flow from the login handler to the database?
Find all places where user permissions are checked.
Trace the request lifecycle from router to response.

Good structural queries

Find async functions that await without try/catch.
Find calls to oldClient.query($$$) outside repository classes.
Find React components that use useEffect and setState.

Plain text queries

Find UserService mentions. → use rg unless AST shape matters.
Search for import React. → use rg.
Find TypeScript files. → use fd.

Workflow

For trace-flow requests, explicitly say you will first check for CONTEXT.md, CONTEXT-MAP.md, nearest */CONTEXT.md, and docs/adr/*.md when present before naming domain concepts.

1. Classify the ask: structural pattern, exact text, file discovery, or semantic flow.
2. Load domain docs when present: CONTEXT.md, CONTEXT-MAP.md, nearest */CONTEXT.md, and relevant ADRs.
3. Run the first local search:
   • Structural: ast-grep run or ast-grep scan.
   • Text: rg -n 'literal|symbol|route'.
   • Files: fd 'auth|login|session|user'.
4. Run semantic code search for multi-hop flow: WarpGrep or another available semantic search tool.
5. Read specific files or line ranges only when needed for verification.
6. Return a bounded code map with file paths and line references.

Zoom-Out Mode

Use when the user says "zoom out", "map this area", "go up a layer", or sounds lost in local details.

Return a map, not a dump:

• relevant modules and callers
• data/control flow across seams
• domain terms from CONTEXT.md
• ADR constraints that shape the design
• known facts vs guesses/unknowns
• where to read next, limited to the top 3 files

Avoid line-by-line explanations unless asked. The point is orientation, not making paste soup.

Final Answer Contract

Return a bounded code map:

1. Flow with file:line references.
2. Key modules and responsibilities.
3. Callers/callees and shared types/messages.
4. Unknowns or unverified assumptions.
5. Read-next list, top 3 files only.

For structural searches, also include the ast-grep pattern or rule used.

Failure Handling

• ast-grep missing: report it and fall back to rg / fd only for the parts they can answer. Suggest brew install ast-grep, npm install -g @ast-grep/cli, or cargo install ast-grep.
• ast-grep no matches: simplify the rule, debug the parsed pattern, verify a target snippet exists with rg, then report the gap without inventing results.
• WarpGrep unavailable: fall back to ast-grep for structural entry points, rg/fd for text/file searches, and Read for specific files; say which tool is being used.
• Request too vague: refuse the full dump, offer a zoom-out map scoped to a specific module or flow, and ask for a narrowing constraint.
• No results found: broaden the query, check alternate naming conventions, or report the gap explicitly. Never fabricate results.