Code Navigation — act101 Agent Skill

Code Navigation with act

Use act's query tools to traverse large repositories efficiently. Do NOT read entire files when you can query for specific information.

Rules

1. Start with repo-outline — Before exploring a codebase, run repo_outline to understand the file tree, languages, and structure. Use --symbols for files of interest.

2. Use skeleton for file structure — When you need to understand a file's structure, use skeleton to see declarations without bodies. Never read an entire file just to find function names.

3. Use interface for API surfaces — When you need to understand how to use a class or module, use interface to get signatures, types, and docstrings without implementation details.

4. Follow the dependency graph — Use graph to understand how files are connected. Start from the file you're interested in with --direction out (what it depends on) or --direction in (what depends on it).

5. Use mutations for raw side-effect detection — Before modifying a function, use mutations to see the individual reads/writes of external state.

6. Use effect_summary for a categorized effect picture — When you want the function's effects classified (reads, writes, raises, allocations, blocking calls, awaits) plus a purity verdict and an unresolved-call frontier, use effect_summary. It is the higher-level view over mutations: reach for effect_summary to answer "what does this touch, and is it pure?" before a change; drop to mutations for the raw access list.

7. Use control-flow for complex logic — When a function is hard to understand, use control_flow to get a linearized view of its branching structure.

8. Use data_flow to trace values through a function — The dual of control_flow: data_flow returns variable definition sites, use sites, def→use edges, and the locals that flow into the return. Reach for it to answer "where does this value come from / where does it go" within one function.

9. Batch symbol retrieval — When you need symbols from multiple files, use symbols_batch --files instead of making separate symbols calls. When you need specific implementations, use symbols_batch --ids with stable IDs.

10. Use stable symbol IDs — After finding a symbol, use its stable ID (format: file::QualifiedName#kind) for subsequent operations. This avoids ambiguity and eliminates the need to specify --file.

11. Use definition for cross-file navigation — When you find a reference to an unknown symbol, use definition to jump to its source.

12. Analyze before modifying — Before making changes, run effect_summary (or mutations) on affected functions and graph on affected files to understand the blast radius.

Token-Saving Hints

• repo_outline costs ~7 tokens/file (vs reading files: ~250 tokens/file)
• skeleton costs ~15 tokens/declaration (vs reading full file)
• interface costs ~25 tokens/member (vs reading implementation)
• Always use compact mode (default) — ranges are strings, not objects
• Use --depth limits on graph and repo_outline to control output size
• Filter symbols_batch with --kinds to get only what you need