majiayu000 Code Navigation
Decision Tree: Serena vs Native Tools
| Task | Use | Not |
|---|---|---|
| Reading code structure | Serena get_symbols_overview → read only needed bodies |
Don't read full files |
| Finding definitions | Serena find_symbol |
Don't use Grep for symbol defs |
| Finding callers | Serena find_referencing_symbols |
Don't use Grep for who-calls-what |
| Replacing functions | Serena replace_symbol_body (entire function) |
Use Edit for line-level changes |
| Non-code files | Read/Edit/Grep | Serena only works with LSP languages |
| Config/markdown/TOML | Read/Edit | Serena has limited non-code support |
Serena Capabilities
Symbol Navigation
find_symbol: Locate symbol definitions across the codebaseget_symbols_overview: Get file structure without reading full content (50-75% token savings)
Impact Analysis
find_referencing_symbols: Find all callers/users of any symbol across the codebase
Precise Edits
replace_symbol_body: Replace entire function/method bodiesinsert_before_symbol: Insert code before a symbol definitioninsert_after_symbol: Insert code after a symbol definition
Rename Refactoring
rename_symbol: Rename symbols with automatic reference updates across entire codebase
Pattern Search
search_for_pattern: Regex search across project files
Project Memory
read_memory: Retrieve project-specific code structure and conventionswrite_memory: Store project-specific code structure and conventions
Progressive Disclosure Pattern
- Start with
get_symbols_overviewfor a file (50-75% token savings vs full read) - Read only the function bodies you actually need with
find_symbol+include_body=True - Use
find_referencing_symbolsto trace callers before modifying any symbol - Fall back to Read/Edit for non-Rust files
Token Budget
get_symbols_overviewsaves 50-75% tokens compared to reading full source files- Always prefer symbol-level reads over full-file reads for Rust code
- A single
find_symbolcall is cheaper than reading and searching a full file