audit-cicd
Warn
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes destructive shell commands using the GitHub CLI ('gh') to delete repository resources via the API. Evidence in Step 4: 'gh api -X DELETE "repos///actions/artifacts/{}"' and 'gh api -X DELETE "repos///actions/caches/{}"'. These commands could lead to unintended data loss if executed without proper validation of targets.
- [PROMPT_INJECTION]: The instructions use restrictive phrasing designed to override the agent's default reasoning and autonomy during execution. Evidence: The use of '[LOW freedom — run exactly]' and '[LOW freedom — do not skip]' as explicit behavioral constraints.
- [COMMAND_EXECUTION]: The provided scripts use shell interpolation for repository names and owners without explicit sanitization steps. Evidence in Step 0 and Step 4: 'gh repo list ' and 'gh api "repos///..."'. This pattern carries a risk of command injection if the agent uses unsanitized user-provided input in a shell environment.
Audit Metadata