commit-changes
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute various git commands (
git status,git diff,git commit) and shell utilities likeawkto analyze the repository state and manage the version control workflow. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data (code diffs and file contents) to generate summaries. While this represents a vulnerability surface for malicious instructions embedded in codebase changes, the skill mitigates this by mandating adherence to the actual file tree and forbidding the description of behaviors not present in the diff.
- Ingestion points: Project source code, git diffs, and git status output via local command execution.
- Boundary markers: Relies on standard Git output formatting; no specific delimiters are defined for the summarized content.
- Capability inventory: Full access to shell commands via
git, file system read/write access for managing commits and message files. - Sanitization: No specific sanitization of diff content is performed, but instructions strictly enforce accuracy and verification against the project's own command-line tools.
Audit Metadata