commit-and-push
Pass
Audited by Gen Agent Trust Hub on May 11, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill interpolates user-controlled data directly into a shell command execution string.\n
- Ingestion points: The 'intent' derived from the conversation history and the Claude model name provided via the
$ARGUMENTSvariable are both used to build the command line forgit-agent commit.\n - Boundary markers: The instructions attempt to wrap the interpolated variables in double quotes (e.g.,
--intent "<intent>"), which provides a basic level of shell escaping.\n - Capability inventory: The skill is granted access to
git-agentandgitbinaries via theallowed-toolsfrontmatter, and it uses these to perform commits and network-basedgit pushoperations.\n - Sanitization: There is no explicit sanitization or filtering of the derived intent or argument strings before they are passed to the shell, which represents a standard command injection surface for this type of tool.
Audit Metadata