cleanup-project
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns detected. The skill demonstrates high security awareness through its handling of sensitive files and shell command construction.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository data such as filenames, diffs, and commit logs. \n
- Ingestion points:
SKILL.md(Step 1, 2, 5) andreferences/uncommitted-review.md(Section 1, 2) read output from git status, diff, and log commands. \n - Boundary markers: Employs structured tables in
action-plans.mdand custom headers like◆ Cleanup Reportto delimit data from instructions. \n - Capability inventory: Includes branch deletion (
git branch -D), file restoration (git restore), and repository cleaning (git clean -f). \n - Sanitization: Uses strict single-quoting for all path arguments and
git --literal-pathspecsto prevent command injection. Every destructive operation requires explicit, consolidated user approval before execution.\n- [COMMAND_EXECUTION]: The skill executes shell commands using git and gh. It explicitly instructs the agent on safe practices to mitigate injection risks from untrusted file paths (e.g., avoiding double quotes and shell expansion).
Audit Metadata