| id | architectural-grounding |
| name | Architectural Grounding (Large File Discovery) |
| description | Safely explores and categorizes large entry-point files (>300 lines) without consuming massive context tokens. |
| tools_used | ["list_symbols_in_file","summarize_code_intent","search_definitions"] |
| trigger_conditions | ["Assigned to a task in a file > 300 lines","exploring \"App\" or \"Server\" structs"] |
Skill: Architectural Grounding
Goal: Understand the responsibility of a complex file without reading the raw source.
Execution Pattern
- Metadata Triage: If a file is known to be large (e.g.,
app.go, main.go, server.go), call list_symbols_in_file with distill: true immediately.
- Component Isolation: Identify the 2-3 specific methods or structs in the distilled summary that are relevant to the query.
- Symbolic Zoom: Call
search_definitions or get_symbol_map on those specific components to find their definitions.
- Surgical Read: ONLY call
read_file or view_file for the specific line ranges identified in step 3. NEVER read more than 100 lines at once.
Success Metric
The agent should be able to explain the "Flow" of a feature (like Database Init) while having read fewer than 15% of the total lines in the file.