auto-push
Pass
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes multiple shell commands to synchronize the repository, including
git fetch,git add .,git commit, andgit push. - [DATA_EXFILTRATION]: Performs network operations via
git pushto send local repository content to a remote server. While it includes safety checks for secrets and API keys, the use ofgit add .risks staging and pushing sensitive information if the filtering logic is bypassed or if a sensitive file is not in the exclusion list. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by processing untrusted data (file names, diff statistics, and commit logs) to generate commit messages and summaries.
- Ingestion points: Data enters the context via
git status,git diff --stat, andgit logoutput inSKILL.md. - Boundary markers: Absent. There are no delimiters to separate untrusted git output from agent instructions.
- Capability inventory: Subprocess execution for git commands and network operations via
git push. - Sanitization: No sanitization or escaping of external repository content before interpolation into prompts.
Audit Metadata