commit-push
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local shell commands including
gitandgh(GitHub CLI) to manage the repository state. These commands are standard for version control tasks and aligned with the skill's purpose. - [INDIRECT_PROMPT_INJECTION]: The skill ingests external content from the local filesystem during analysis, which could theoretically contain malicious instructions designed to influence the agent's behavior during the commit process.
- Ingestion points:
git diff HEADandgit log --oneline -5are used to gather context from the current workspace and history (SKILL.md). - Boundary markers: The instructions rely on structured phases but do not use specific XML or text delimiters to isolate the code being analyzed from the agent's instructions.
- Capability inventory: The skill has the ability to stage files (
git add), create commits (git commit), and push changes to a remote server (git push). - Sanitization: The skill implements strong human-in-the-loop controls, requiring user confirmation of the scope (Phase 1) and the commit contents (Phase 4) before any network operations occur.
Audit Metadata