interrogate-codebase

Load the codebase before working on a feature

Before touching existing code, a senior developer doesn't just dive in — they reopen every file related to the feature to refresh their mental model. This skill makes that warm-up mandatory and exhaustive so the agent works from real loaded context instead of guessing from memory.

Use when

• About to implement, fix, refactor, or extend a feature that already has code (frontend, backend, or both).
• The user names a scope: "the search algorithm", "the cart flow", "the export job".

Process

1. Pin down the scope. The user's prompt names a feature, area, or bug. If the boundary is genuinely ambiguous, ask once. Otherwise trust the prompt — the responsibility to be precise lies with the user.

2. Identify every file in scope, across both layers.

   • Frontend: components, pages, hooks, state/store, client routes
   • Backend: routes/controllers, services, models, jobs, crons, migrations, configs
   • Include shared utilities, types, and helpers touched by the scope
   • Exclude tests — devs don't re-open tests when refreshing memory on a feature
   • Exclude vendored code, node_modules, vendor/, build artifacts

3. Read each identified file in full. Use the Read tool on every file. No partial reads, no excerpts — the agent needs the same context a developer would have after opening them.

4. Stop. Do not write a summary, a map, an index, or a "files loaded" recap. The Read calls already put the files in context. Move directly to the user's task.

Interaction with other skills

• Run BEFORE investigate-ui-first — code grounding first so the UI investigation knows which surfaces to sample.
• Run AFTER repo-onboard if you've never touched this repo — that one gives the macro picture; this one zooms into the scope at hand.
• Run BEFORE Superpowers brainstorming and writing-plans — design and plan should be informed by the actual loaded code, not guessed convention.