The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches source code and configuration files from an external GitHub repository (github.com/cnitlrt/AutoTeam) that is not part of the trusted vendor list.
[REMOTE_CODE_EXECUTION]: The installation process involves downloading a repository and immediately executing a shell script (bash setup.sh), allowing for the execution of unverified remote code on the system.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI tools and subprocesses, including uv for environment management, playwright for browser automation, and docker compose for service orchestration.
[CREDENTIALS_UNSAFE]: The primary function of the skill involves managing and synchronizing sensitive credentials, including TEAM_ADMIN_PASSWORD, CPA_API_KEY, and CLOUDMAIL_API_KEY, which are stored in environment files and an SQLite database.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The browser automation logic in src/autoteam/browser.py reads and processes verification content from external temporary email services.
Boundary markers: No explicit boundary markers or isolation protocols are defined for handling the untrusted text from email bodies.
Capability inventory: The skill possesses significant capabilities, including full browser control, network exfiltration potential via httpx, and filesystem write access for configuration and database files.
Sanitization: There is no documented evidence of sanitization or validation performed on data ingested from external mail APIs before it influences automation workflows.

autoteam-chatgpt-rotation