kabeuchi
Pass
Audited by Gen Agent Trust Hub on Aug 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill registers a
UserPromptSubmithook that executes a shell script (prompt-hook.sh) across several potential installation paths. It implements significant security measures, including a content-based verification step (checking for a marker string) and a specific search order that prioritizes user-controlled directories over project-local ones to mitigate risks from malicious files in a cloned repository.\n- [COMMAND_EXECUTION]: The skill utilizes dynamic context injection inSKILL.mdto initialize session markers via pre-execution shell commands. The implementation demonstrates security awareness by explicitly avoiding the use of user-provided arguments in these commands, thereby preventing shell injection vulnerabilities at skill load time.\n- [PROMPT_INJECTION]: The skill processes untrusted data from local markdown files and GitHub issue bodies, which is then incorporated into the agent's context during the refinement process. This creates a surface for indirect prompt injection if the source content contains malicious instructions.\n- [EXTERNAL_DOWNLOADS]: The documentation identifies the/grillingskill as a required dependency and provides a link to its official GitHub repository. This reference targets a well-known service and requires manual user intervention for installation, posing no automated risk.
Audit Metadata