comet-verify
Warn
Audited by Gen Agent Trust Hub on Jun 14, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a wide array of system commands and bash scripts to manage state, interact with Git (git diff, git stat), and perform file system operations (mkdir, grep, find).
- [REMOTE_CODE_EXECUTION]: In Step 0b, the skill uses the find command to locate a script named comet-env.sh across multiple paths, including $HOME/.config and $HOME/.gemini. It then sources this script using the shell's dot operator ('. "$COMET_ENV"'), which executes the content of the discovered file in the current execution context. This constitutes dynamic loading and execution from a path determined at runtime.
- [DATA_EXFILTRATION]: The script accesses sensitive directory paths such as $HOME/.config and $HOME/.gemini during its discovery process. While the intent is to find its own environment script, this demonstrates broad access to sensitive configuration areas.
- [PROMPT_INJECTION]: The skill's logic is heavily driven by the content of project-level files like tasks.md, design.md, and proposal.md. It parses these files to determine verification criteria and task completion, which represents an indirect prompt injection surface if those files are modified to include instructions that subvert the verification process.
Audit Metadata