commit-work

Pass

Audited by Gen Agent Trust Hub on May 10, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface. It extracts information from local proposal files (openspec/changes/<id>/proposal.md) to determine the commit title. There are no boundary markers or sanitization steps to ensure that instructions embedded in the proposal file do not influence the agent's behavior.
  • [COMMAND_EXECUTION]: The skill uses potentially untrusted data from proposal files in a bash command (git commit -m "<title>"). If a proposal title contains shell metacharacters such as backticks, semicolons, or quotes, it could lead to arbitrary command execution during the commit process.
Audit Metadata
Risk Level
SAFE
Analyzed
May 10, 2026, 01:35 PM