tool
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill retrieves and executes shell commands from external 'recipe' files provided by a central catalog or local directories. This mechanism allows for the runtime execution of code defined outside the skill's own package.
- [DYNAMIC_EXECUTION]: The skill dynamically parses Markdown files using a Bash script to extract and execute fenced shell code blocks. This behavior transforms static data (recipes) into executable commands.
- [INDIRECT_PROMPT_INJECTION]: The skill processes project-level configuration files (
project.json,alignment.json,task.json) and external Markdown recipes. These files represent an attack surface where maliciously crafted external data could influence the agent's behavior and the commands it executes. - [COMMAND_EXECUTION]: The skill uses a dedicated Bash script (
tool-actions.sh) to execute shell commands. While it limits execution to a single command for tool runs and attempts to block shell metacharacters, it remains a powerful capability for interacting with the host system.
Audit Metadata