The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python script scripts/verifier.py to perform resource verification. The instructions recommend using python scripts/verifier.py --text "<input_text>", which creates a significant surface for shell command injection if user-supplied text is interpolated directly into the command line without proper escaping. The author provides a mitigation strategy (saving to a temporary file), acknowledging this risk.
[COMMAND_EXECUTION]: The scripts/verifier.py script explicitly disables SSL certificate verification (verify=False) in both its synchronous and asynchronous httpx clients. This insecure configuration makes the verification process vulnerable to Man-in-the-Middle (MITM) attacks when the agent connects to external URLs or registries.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from user input, local files, and fetched URLs.
Ingestion points: User-pasted text, local file paths provided by the user, and content retrieved via webfetch from user-supplied URLs.
Boundary markers: None implemented in the script execution phase to separate user content from command arguments.
Capability inventory: Subprocess execution of Python scripts, file system read access, and arbitrary network operations via httpx and webfetch tools.
Sanitization: The script uses regular expressions to extract links and package names, but it does not validate or sanitize the target URLs before initiating network requests, creating a surface for Server-Side Request Forgery (SSRF).
[EXTERNAL_DOWNLOADS]: The skill performs network requests to verify the existence and accessibility of resources. It targets well-known services including the PyPI registry (pypi.org), the NPM registry (npmjs.com), and the DOI resolver (doi.org). While these are trusted sources, the script also attempts to visit any arbitrary URL extracted from the input text.

ai-content-verifier