Logic Flow Skill | Agent Skills

Logic Flow

Explain the logic flow of a function by examining its implementation, control flow, and callees.

Process

Step 0: Orient (Optional)

If the function is in an unfamiliar area, get context first:

get_context(query="<function_name or related concept>")

This shows the function's neighbors and where it fits in the broader architecture.

Step 1: Deep Dive with Full Depth

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

This returns the function's code body, callers, callees, and type information.

If deep_dive returns the wrong symbol (common names), use context_file to disambiguate:

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

Step 2: Analyze the Code

Read through the function body and identify:

Entry: Function signature, parameters, what they represent
Guard clauses: Early returns, validation, error checks at the top
Setup: Variable initialization, resource acquisition
Core logic: The main processing — what does this function actually DO?
Control flow: Branches (if/else, match), loops, error handling (? operator, try/catch)
Callees: What other functions are called and why
Return: What value is produced and what it means

Step 3: Write Step-by-Step Explanation

Function: <name> (<file>:<line>)
Purpose: <1-sentence summary of what this function does>

Parameters:
  - param1: Type — what it represents
  - param2: Type — what it represents

Logic Flow:
  1. Validates input: checks X and Y, returns early with error if invalid
  2. Acquires database lock via get_connection()
  3. Fetches matching symbols from database (calls fetch_symbols)
  4. If query is provided:
     a. Builds Tantivy search query
     b. Searches index for matches (calls search_index)
     c. Ranks results by BM25 score
  5. Otherwise: returns all symbols sorted by name
  6. Formats results into CallToolResult and returns

Key Callees:
  - fetch_symbols(): Retrieves symbols from SQLite by workspace ID
  - search_index(): Runs Tantivy full-text search with code tokenizer

Error Handling:
  - Returns Err if workspace not initialized
  - Logs warning and continues if individual file parsing fails

Returns: Result<CallToolResult> — MCP response with formatted search results

Important Notes

Don't just repeat the code — explain the WHY, not just the WHAT
Focus on the interesting parts — skip boilerplate, highlight decisions
For large functions (>100 lines): Summarize sections rather than explaining every line
For trait methods: Note which implementations exist and which is being analyzed
Reference workspaces: Pass workspace: "<workspace_id>" to all tool calls when analyzing a non-primary workspace

Agent Skills: Logic Flow

Install this agent skill to your local