The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/media_proxy.py reads user credentials from the sensitive file path ~/.gooseworks/credentials.json to obtain API keys and service endpoints.
[DATA_EXFILTRATION]: Sensitive API tokens are transmitted as query parameters in HTTP requests (e.g., ?token=...). This is an insecure pattern as URL parameters are commonly logged in plain text by proxy servers, web servers, and browser history, leading to potential credential exposure.
[DATA_EXFILTRATION]: The skill provides the ability to write external content to arbitrary local paths via the download function and the resume.py script's --out argument. Using urllib.request.urlretrieve and pathlib.Path.write_bytes, it lacks validation on the destination path, creating a risk where the agent could be manipulated into overwriting critical system files or configurations (e.g., ~/.ssh/authorized_keys).
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted data:
Ingestion points: Arguments such as payload, prompt, and url in scripts/media_proxy.py are interpolated into network requests or used for downloads.
Boundary markers: There are no markers or instructions to help the agent distinguish between data and potentially malicious embedded commands.
Capability inventory: The skill possesses powerful capabilities including network requests via requests and file system writes via urllib.request and pathlib.
Sanitization: There is no evidence of input validation, filtering, or escaping for the external content processed by the skill.

media-proxy