What problem does it solve? After running automated PR reviews, you may find that review agents produce false positives, miss real issues, or review files outside the PR diff. This Skill closes that feedback loop by turning a Prompt Improvement Assessment into concrete, validated edits to review prompts, the TypeScript orchestrator, or review configuration. ## Core Features & Use Cases - Assessment-driven prompt edits: Parses a Prompt Improvement Assessment into classified action items (prompt-edit, orchestrator-edit, config-edit, no-action) and applies minimal, targeted changes to prompts in global/prompts/{claude,codex,gemini}/. - Orchestrator and config fixes: Makes surgical edits to tools/stark_review.ts (with tests for logic changes) and adds backward-compatible config fields like disabled_paths or severity_overrides. - Validation and changelog: Runs syntax checks, focused tests, and JSON validation, then records every change in a prompt changelog with source, reason, and validation status. - Use Case: After a review run flags v2 concerns as v1 issues, use this Skill to add a scope calibration instruction to the relevant domain prompt, validate the change, and log it — committing only when you explicitly approve the diff. ## Quick Start Ask the assistant to run stark-review-improvement on the prompt improvement assessment from your latest review to fix the noisy review prompts.
