Matt Pocock's skills add /review workflow

The useful part here is the shift in posture: the agent stops acting like an implementation engine and starts acting like a reviewer that can call out process failures too. That’s a more honest workflow for AI coding, where the biggest misses are usually misalignment, missing tests, and too much unexamined code churn.

• –Good review output is opinionated and failure-focused: spec drift, standard violations, regressions, and missing checks come first
• –Asking the skill to propose changes to the agent loop is the right move; many code quality problems are workflow bugs, not just code bugs
• –This fits Matt Pocock’s broader pattern: small, composable skills that encode engineering habits instead of giant catch-all prompts
• –The main risk is over-trusting the reviewer agent; human judgment still needs to own ambiguous tradeoffs and architecture calls
• –If paired with TDD and planning skills, this becomes a practical guardrail stack for AI-assisted development