Type Flow Skill | Agent Skills

Type Flow

Trace how types flow through a function by analyzing type signatures, transformations, and conversions.

Process

Step 1: Deep Dive the Function

deep_dive(symbol="<function>", depth="full")

Extract the function signature, body, and type information.

If deep_dive returns the wrong symbol, use context_file to disambiguate:

deep_dive(symbol="<function>", depth="full", context_file="<partial_file_path>")

Step 2: Map the Type Pipeline

Trace each type from input to output:

Input types: What are the parameter types?
Transformations: Where does a type convert to another?
- Conversion methods (Rust: .into(), From::from(); JS/TS: type assertions; Python: constructors)
- Method calls that return different types
- Destructuring / pattern matching / spread operators
Intermediate types: What types exist between input and output?
Output type: What does the function return?

Step 3: Find Type Definitions

For any non-obvious type, look it up with a quick definition search:

fast_search(query="<TypeName>", search_target="definitions")

Or for a deeper look at the type's structure:

deep_dive(symbol="<TypeName>", depth="overview")

To see how a type is used across the codebase:

fast_refs(symbol="<TypeName>", reference_kind="type_usage", limit=10)

Step 4: Report

Function: <name>

Type Flow:

  Input:
    self: &JulieServerHandler
    params: FastSearchTool { query: String, limit: Option<u32> }

  Pipeline:
    1. FastSearchTool → extract query: String, limit: Option<u32>
    2. String → tantivy::Query (via QueryParser::parse_query)
    3. Query → Vec<(DocAddress, f32)> (via Searcher::search)
    4. DocAddress → TantivyDocument (via Searcher::doc)
    5. TantivyDocument → Symbol (via database lookup)
    6. Vec<Symbol> → String (via format_results)
    7. String → CallToolResult (via CallToolResult::text_content)

  Output:
    Result<CallToolResult> — MCP response wrapper

  Key Type Definitions:
    - FastSearchTool: src/tools/search.rs:15 (tool parameter struct)
    - CallToolResult: rmcp crate (MCP protocol response)
    - Symbol: src/extractors/mod.rs (code symbol representation)

  Complexity: Medium (4 type transformations)

Important Notes

Follow error propagation — Rust's ?, JS/TS try/catch, Python raise — these change the effective type in the success path
Generic types: Note type parameters (e.g., Vec<T> where T matters)
Trait objects: dyn Trait or impl Trait — note what concrete types are used
Closures: Types of closure parameters are often inferred, state what they are
Reference workspaces: Pass workspace: "<workspace_id>" to all tool calls when analyzing a non-primary workspace

Agent Skills: Type Flow

Install this agent skill to your local