Security audit
Security checks for vulnerabilities and agentic risk
This review skill is not malicious, but it can automatically edit repository files during broadly triggered code review sessions without a separate confirmation step.
Install only if you are comfortable with a review skill that may edit files to fix issues it considers clear. Review local diffs after each run, and avoid invoking it casually unless you want possible code changes as well as review feedback.
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
- Do not commit, push, or open a PR.
- Do not mix review with broad refactoring.
## Output Rules
- `HARD STOPS` must appear first.
- Every hard stop needs a file/line or artifact reference when available.
The trigger list contains broad phrases like 'check this', 'look for problems', and pre-merge situations that can match normal conversation too easily. Because this skill has write-capable tools and auto-fix instructions, overbroad invocation materially raises the risk of unintended execution and repository modification.
This review skill is presented as an analysis gate, but it explicitly authorizes directly editing repository files to auto-fix findings. That creates a privilege expansion from passive review into active code modification, which can silently alter code under the guise of review and increases the chance of unauthorized or unsafe changes being introduced.
The skill's stated purpose is reviewing diffs, but it includes code-editing capability and instructions to change files when a fix seems clear. That violates least privilege for a reviewer and makes accidental or adversarial modification more likely, especially when broad trigger phrases may invoke the skill in routine conversations.
The trigger list hard-codes English and Chinese phrases as activation language cues, but does not state whether users may interact in other languages or whether the skill is limited to those locales. This can be interpreted as a language/locale constraint without explicit opt-in or justification.