--- name: exploring-code description: Use when the user asks how a feature or flow works, where something is implemented, or to walk through code end to end. Skip when the user names the file to read or the text to find. Load this skill before the first tool call. metadata: opencode/slash: "true" --- # Exploring code Choose the evidence the question needs. Literal text search finds strings and configuration keys; symbol lookup finds declarations; concept search helps when the implementation's name is unknown. Text matches do not establish call edges. Known identifiers make useful lexical anchors; preserve them when continuing a search instead of broadening an already specific question. Resolve an ambiguous symbol to its file and node before reading its body or following relationships. An outline helps locate a body in an unfamiliar file; a direct bounded read is sufficient when the path and relevant region are known. Read response freshness and coverage with the result. Missing or partial indexed coverage cannot prove absence. Generation-bound code navigation belongs to the exact project and worktree snapshot; do not substitute the active graph for an unavailable requested scope. Continuations retain the original lexical anchors, preferences, and opaque cursor; a typed mismatch means resolve the query again. For type/trait questions, see [types and traits](references/types-and-traits.md). For cross-branch reads, see [other branches](references/other-branches.md). Call relationships belong to `tracing-functions`; changed consumers and tests belong to `assessing-impact`. **Report:** the answer with the file and symbol behind each claim, plus the freshness line when it is not `fresh` and the coverage when it is partial. Do not claim absence from an empty or partial result.