The Agent Skills Directory

[SAFE]: The skill serves as a template and instruction set for a PR verification agent (qwen-code). It emphasizes objective measurement (A/B testing, wire oracles) over subjective impressions.
[SAFE]: The frontmatter and instructions correctly identify that the skill is intended to run in a 'credential-free container or VM' with no access to host secrets (SSH keys, cloud profiles, or tokens). It explicitly warns against running it in an ordinary working copy on a maintainer's machine.
[SAFE]: Network operations are restricted; the skill instructions state that the environment should 'make no network calls' and treat the local snapshot as the 'whole world.' This is a strong security posture for a sandbox environment.
[SAFE]: The execution of PR code is recognized as an intended function within a designated sandbox, and the instructions provide rigorous methodologies for ensuring that verification is load-bearing and not vacuous (e.g., mutation testing, A/B counts).
[SAFE]: The references to qwenlm resources (like qwen-triage.yml and @qwen-code packages) are internal vendor resources consistent with the skill's stated purpose.

verify-pr