principle-test-behavior-not-implementation

Explore a codebase and return an evidence-backed map.

Create a repository-local Claude Code skill that verifies an engineering workflow through its public entry point and observable results.

Review and commit the currently staged Git changes.

Multitask across asynchronous subagents.

Review only the files changed in the current task by pinning a temporary Git snapshot and delegating independent standards and spec reviews.

Create a redacted, portable handoff document for continuing work in a later Claude Code session.

Reconstruct working context from an explicit handoff, decision log, Git or PR state, current-chat context, and user-provided evidence.

Disciplined diagnosis loop for hard bugs and performance regressions.

Sketch types, signatures, and module structure before code, then stay in the loop while implementation fills in.

Shared vocabulary for designing deep modules.

Find what a change could break somewhere else before it ships, beyond the diff, and prove the one fact it's safe because of by running real code instead of…

Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it.