| name | skill-safety-audit |
| description | Scan new or updated skills for unsafe or malicious instructions (unknown tools, external installers, credential harvesting) before accepting them into the repository. |
Skill Safety Audit
Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, +256 784 464178.
Use when
- The task matches this domain: Scan new or updated skills for unsafe or malicious instructions (unknown tools, external installers, credential harvesting) before accepting them into the repository.
- The user needs evaluation, findings, prioritization, or remediation guidance.
Do not use when
- The user needs direct implementation instead of an audit.
- There is no target artifact or change set to inspect.
Required inputs
- The site, page, skill, or workflow being audited.
- Any business goals, constraints, or supporting evidence available for prioritization.
Workflow
- Define the audit scope and what success or risk means in context.
- Inspect the target using the relevant checklist and preserved guidance.
- Summarize the highest-value findings with clear reasoning.
- Prioritize remediation so the next actions are obvious.
Quality standards
- Findings must be evidence-based and ordered by impact.
- Recommendations should be specific enough to implement.
- The audit should reduce risk or ambiguity for the next person acting on it.
Anti-patterns
- Do not provide generic checklist output with no prioritization.
- Do not hide high-severity issues in vague language.
- Do not imply certainty when evidence is missing.
Outputs
- Audit report, prioritized findings, remediation notes, or approval status.
References
- Start with
references/legacy-guidance.md when you need the preserved detailed instructions from the previous skill version.
- Read only the specific files under
references/ that match the current task instead of loading the whole directory.
- This skill has no bundled scripts by default; keep execution focused on the documented workflow and any existing project files.
Notes
- Treat this
SKILL.md as the portable execution layer for both Codex and Codex.
- Preserve existing project behavior unless the current task explicitly requires a change.