The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/start_login.py uses subprocess.Popen to open the system's default web browser for authentication. On Windows, this is implemented using shell=True. While the URL is constructed from a hardcoded base and a session ID from the vendor's API, the use of the shell for command execution is noted as a standard but sensitive operation.
[EXTERNAL_DOWNLOADS]: The skill fetches platform data, such as basket details and market analytics, from the vendor's official backend at https://backend.cesto.co. These operations are limited to the vendor's infrastructure and are used solely to present data to the user.
[SAFE]: The skill implements several robust security practices:
Session Isolation: Sensitive authentication tokens are stored in an encoded format in ~/.cesto/session.dat and are managed by helper scripts. This ensures the AI model never handles raw secrets, preventing credential leakage in conversation logs.
Indirect Prompt Injection Defenses: The SKILL.md file contains explicit security instructions directing the agent to treat API content as untrusted user-generated data. It specifically forbids the agent from following URLs or executing instructions found within basket descriptions.
URL Allowlisting: The scripts/api_request.py utility enforces a strict allowlist of origins, ensuring that authentication headers are only sent to the vendor's authorized domains.

cesto-toolkit