commit

Warn

Audited by Gen Agent Trust Hub on Feb 25, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The Troubleshooting section includes an instruction to execute echo $GPG_PASSPHRASE. If an agent follows this logic during a failure state, it would output the plaintext secret to the console or log files, resulting in credential exposure.
  • [COMMAND_EXECUTION]: The skill uses bash to interact with git and gpg. It specifically utilizes process substitution (<(echo "$GPG_PASSPHRASE")) within a shell script to pipe sensitive environment variables into the GPG binary for credential caching.
  • [PROMPT_INJECTION]: This skill is vulnerable to indirect prompt injection.
  • Ingestion points: Reads repository state and changes via git status and git diff (SKILL.md).
  • Boundary markers: None identified; diff content is processed directly to generate commit messages and logic.
  • Capability inventory: Execution of git commit and arbitrary bash commands (SKILL.md).
  • Sanitization: No validation or escaping of the diff content is performed before it is used to influence the agent's natural language output and command generation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 25, 2026, 11:50 AM