The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md include multiple shell command blocks for verification (e.g., in Section 6 and 7) that use the <tool> variable. If a user provides a tool name like 'tsc; malicious_command', the agent may execute the unauthorized command on the host.
[COMMAND_EXECUTION]: Command templates such as uv run pytest tests/unit/tools/<tool>/, npm view <tool> version, and uv run lintro check . --tools <tool> are vulnerable to injection because there are no boundary markers or instructions to sanitize the tool name input before interpolation into shell commands.

lintro-verify