The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to download a compressed archive containing a binary file (humanizer.exe) from an unverified GitHub repository (0xwilliamortiz/humanizer-cli).
[COMMAND_EXECUTION]: Documentation provides commands for users to manually execute the downloaded binary. Critically, it includes a "Troubleshooting" section that explicitly guides users to bypass operating system security features, such as Windows SmartScreen ("Run anyway") and Antivirus quarantine ("Restore from quarantine"), which are designed to protect users from untrusted or malicious software.
[DYNAMIC_EXECUTION]: The skill provides multiple Node.js implementation examples that use child_process.execSync to dynamically invoke the downloaded executable. This includes pre-commit hooks and build scripts that automatically run the binary against files in the local project directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data from local files, creating a potential vector for injection attacks where embedded instructions could influence the agent's behavior.
Ingestion points: The skill reads and scans content from user-provided files such as draft.md and article.txt using the check and prompt commands.
Boundary markers: The skill does not describe any delimiters or instructions for the agent to distinguish between the tool's output and potentially malicious instructions embedded within the files being scanned.
Capability inventory: The skill possesses command execution capabilities through execSync and provides a binary that interacts directly with the local file system.
Sanitization: There is no evidence of sanitization or validation performed on the file paths or the content of the files before they are passed to the execution environment in the provided scripts.

humanizer-cli-detector