Quality

Check Cache Bugs audits a Claude Code setup for cache and cost bugs discovered in March-April 2026, including sentinel, resume/continue, and attribution header issues. Use it to verify your installation isn't affected by known caching defects.

Codebase Health Audit scores a project across seven health categories—secrets, security, dependencies, structure, tests, and more—and produces a prioritized progression plan. Use it to find weak spots and plan improvements.

Self-Improving Skill Review evaluates claims about self-improving Claude Code skills, verifies patterns, and documents proven alternatives like the Claude Reflect System. Use it to research and validate skill improvement mechanisms before adding them to a guide.

Source Command Sync checks synchronization status between a guide repository and its landing site. It runs a sync verification script, formats results, and provides actionable fixes when mismatches are detected.

Evaluate runs a three-stage verification pipeline—mechanical checks, semantic assessment, and optional multi-model consensus—on an execution session or artifact, allowing rejected outcomes to continue through a Ralph convergence chain.

Quality Assessment evaluates any artifact—code, documents, API responses, test output, or custom content—and returns a score with a PASS, REVISE, or FAIL verdict plus actionable suggestions. Use it when you need a fast single-pass quality gate before proceeding or iterating on work.

Next QA runs one unattended iteration of the quality-assurance loop: stewards in-flight QA PRs and builds one queued qa issue on a branch off auto-qa, opening a PR that squash-merges on green CI. Adds tests and tooling only; never edits product source. Use for 'QAタスク', 'next qa', or quality-track progress.

Next QA Idea scans a codebase for the highest-value untested behavior, judges it against a QA value bar, and files one locked `qa` issue specifying the test to write. Use it when the user asks for QA ideation or to refill the QA backlog without implementation.

PR Review Analysis fetches review comments from a pull request, verifies each finding against the actual codebase, and reports validity with recommended actions. It can also suggest replies to each comment after fixes.

Antinet Security Scan performs security and compliance checks on incoming files or URLs and returns a pass/reject verdict together with a scan report. Use it as a mandatory front-line gate before any document processing.

Create Skill Test scaffolds evaluation specs (eval.yaml) for agent skills in the dotnet/skills repository, covering stimuli, graders, rubrics, fixture files, and statistical sizing to avoid overfitting. Use it when creating skill tests or adding evaluation coverage.

Improve Skill Quality diagnoses failing or unconvincing skill evaluations and turns them into targeted fixes. Use it when an evaluation verdict is a regression, underpowered, or returns no credible improvement, or when deciding whether to strengthen or retire a weak skill.