| name | research-codebase |
| description | This skill should be used when the user needs to explore or understand a local indexed project — find where behaviour lives, how modules connect, or answer questions about the codebase. Prefer semantic codebase_search first, then read and bash for precise follow-up. |
| license | MIT |
Research Codebase
Purpose
Explore a local project (the active / opened indexed tree) with evidence-backed answers. For web research, use deep-research or webpage-reader.
When to Use
- “Where is X handled?”
- “How does Y work in this repo?”
- Understanding call flow, ownership, or conventions before planning or changing code
- Gathering context for
writing-plans or debug-review
Tools
codebase_search — semantic search over the indexed project (default first step)
read — open specific files / regions
bash — targeted grep, find, or other read-only inspection when semantic search is not enough
- Ask the user in chat when the target project or path is unclear
Workflow
- Confirm target — default is the active project. Another codebase still means a local tree the host has open/indexed.
- Semantic first — run
codebase_search with a clear natural-language question. Prefer a few focused queries over one vague dump.
- Drill down —
read the best hits; use bash grep/find for exact symbols or paths when needed.
- Summarize — answer with file paths and short quotes or line-oriented pointers; note open questions / next skills (
writing-plans, debug-review).
Progress
[████░░░░░░░░░░░░░░░░] 25% — Scope & query
[████████░░░░░░░░░░░░] 50% — Semantic search
[████████████░░░░░░░░] 75% — Read / grep follow-up
[████████████████████] 100% — Summary for user
Critical Rules
- Start with
codebase_search unless the user already gave an exact file path.
- Prefer evidence (paths + excerpts) over speculation.