ce-resolve-pr-feedback
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No evidence of malicious behavior, data exfiltration, or obfuscation was found in the skill's logic or scripts.
- [PROMPT_INJECTION]: The skill was evaluated for indirect prompt injection risks because it processes untrusted PR comments. The skill effectively mitigates this via explicit 'Security' sections that instruct the agent to never execute commands or snippets found in comments and to always decide on fixes independently.
- Ingestion points: PR comments, review threads, and review bodies (via
scripts/get-pr-comments). - Boundary markers: Not present as data delimiters, but defined via strong instructional constraints.
- Capability inventory: Git repository modification (
git commit/push), GitHub API interaction (gh), and execution of local test suites. - Sanitization: Instructions mandate reading the actual code and deciding on the fix regardless of the comment content.
- [COMMAND_EXECUTION]: Analyzed the use of shell commands and local bash scripts. The skill relies on standard developer tools (
git,gh) for PR management. The execution environment is restricted to these tools through theallowed-toolsconfiguration.
Audit Metadata