Answer structural questions before touching code
Forces the agent to map callers, blast radius, and contracts before editing an unfamiliar symbol — with whatever code-intelligence it has.
by @kourosh · Aug 4, 2026
Fill the placeholders
Before editing {{symbol_or_file}}, answer these structural questions using the best code-intelligence available to you (language server, code graph, repo search — in that order of preference). Do not open random files until these are answered.
1. WHO DEPENDS ON IT — list every caller/importer of {{symbol_or_file}}. If your tool reports truncated results, widen the query; never act on a truncated list.
2. BLAST RADIUS — if its behavior or signature changes, which modules/tests/consumers break, up to two levels out?
3. CONTRACTS — which of its inputs/outputs are load-bearing promises (API responses, event shapes, CLI flags, DB rows) that something else parses?
4. HIDDEN EDGES — any dynamic dispatch, reflection, string-keyed lookup, or config-driven wiring that plain text search would miss?
Only after writing the four answers: state the smallest change that achieves the goal, list the files you will actually open (should be few), and proceed.
Rule of thumb: "where is this?" is a search question; "who depends on this?" is the question that prevents incidents. Always answer the second before editing.
Goal of the change:
{{change_goal}}