Simplify

One review skill for selected changes or an existing codebase. Cursor's Thermo-Nuclear Code Quality Review supplies the structural standards; its rubric below is preserved unchanged. See the included MIT licence.

Scope and actions

• simplify reviews the selected diff, branch or PR. State the comparison base and include relevant surrounding code, dependencies and callers. If no scope is selected, use current uncommitted changes, otherwise branch changes against the repository's default branch. If there are no changes, say so; do not silently review the entire repository.
• simplify whole codebase reviews the current repository, including unchanged code. A path narrows either scope: simplify in <path> or simplify whole codebase in <path>.
• A focus such as security or performance directs attention within that scope. Explicit restrictions win. with scores adds scoring; without it, return findings without a scorecard. These are natural-language arguments, not harness-specific flags.

Both scopes report findings in conversation by default. A review does not authorize code edits, tracker writes, commits or publication. Apply the fixes authorizes implementation within the selected scope; create Linear items authorizes tracker work. Honor existing authorization without asking again. Do not write review artifacts into the repository unless requested.

Use the active harness's tools. No specific harness, integration or model is required. Repository instructions and explicit user constraints govern the review.

Review the code

State the repository, revision, working-tree state and scope. Establish the project's stage using stage guidance; describe an inference as such. Stage affects impact and production-hardening expectations, but never suppresses structural simplification or weakens Cursor's approval bar.

For a whole codebase, map authored source roots and ownership boundaries, excluding dependencies, generated output and vendored code. Examine each in-scope area and trace contracts and data flow between them. For a diff, trace affected paths beyond the changed lines and distinguish new regressions from existing defects relevant to the change. Never present an unrelated pre-existing defect as introduced by the diff.