verify
Warn
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill identifies and executes shell scripts defined in the repository's configuration files (e.g., package.json, pyproject.toml, Makefile, and CI configuration files) using the bash shell. Because the agent is instructed to run these scripts from the repository root, it may execute malicious code if a user or an attacker has modified these configuration files.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content and git diffs through a subagent to perform audits, creating a significant attack surface for indirect prompt injection.
- Ingestion points: Repository configuration files and output from
git diffcommands. - Boundary markers: Absent; the skill does not include instructions to ignore or treat the data as non-executable text, nor does it use delimiters to isolate the diff content.
- Capability inventory: The skill can execute shell commands via bash, read any file in the repository, and invoke subagents with custom prompts.
- Sanitization: The skill lacks sanitization or validation logic to filter out potentially malicious instructions embedded within the code being audited or the repository scripts.
- [DYNAMIC_EXECUTION]: The skill dynamically determines the configuration (model and effort levels) for its auditing subagents by reading JSON files (
.agents/models.json) from either the repository or the user's home directory. This allows the behavior and resource usage of the subagent to be controlled by external configuration files.
Audit Metadata