review-anvil-improve-pr

SUSPICIOUS: the skill's capabilities are largely coherent with its purpose as a PR-improvement workflow, but it is high impact by design: it reads untrusted PR content, modifies local code, pushes to GitHub, and relies on transitive installation of sibling skills from a personal repo. No clear credential harvesting, hidden exfiltration, or off-purpose behavior is present, so this is not malware; the main concerns are prompt-injection exposure and trust in the broader skill chain.