commit
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes Git commands locally to interact with the repository state. Evidence: Run
git status,git diff --staged,git reset,git add, andgit commitinstructions inSKILL.md. - [PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface as it processes untrusted data from the Git index.
- Ingestion points: Reads staged code changes via
git diff --stagedinSKILL.md. - Boundary markers: The skill explicitly mandates a human-in-the-loop checkpoint: 'Only commit after user says yes'.
- Capability inventory: Performs file system modifications through Git index updates and commit creation.
- Sanitization: Relies on user verification of the proposed commit messages and groups to prevent the execution of embedded instructions.
Audit Metadata