The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a repository from an unverified GitHub account (https://github.com/Yurakonoplya/unmark-checker). The repository and its author do not belong to a recognized trusted organization or well-known service.
[REMOTE_CODE_EXECUTION]: The installation instructions involve running pip install -e . on the cloned repository, which executes the repository's setup scripts. Subsequently, the skill executes a CLI tool (unmark-checker) provided by this unverified external source.
[COMMAND_EXECUTION]: The skill provides instructions for executing multiple shell commands, including Git operations, package installation, and the execution of the third-party binary, which grants the software significant control over the local environment.
[CREDENTIALS_UNSAFE]: The skill requires the user to input a secret key (UNMARK_CHECKER_KEY). While the instructions correctly advise against logging the key in shell history, the secret is ultimately passed to and consumed by the unverified external tool.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process text outputs from external "humanizer" or "remover" tools, creating an ingestion point for untrusted data.
Ingestion points: Text files (e.g., cleaned.txt) containing output from third-party services.
Boundary markers: None are specified for the input text.
Capability inventory: File read/write operations and shell command execution via the unmark-checker tool.
Sanitization: The instructions do not mention sanitization or validation of the ingested text before it is processed by the measurement logic.

verify-watermark-removal