The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external websites and applications which could contain malicious instructions. It mitigates this risk by explicitly instructing the agent to treat all tool outputs as data and never as instructions.
Ingestion points: The skill reads website and app inventories via the list_websites and list_installed_apps tools defined in references/tools.md.
Boundary markers: The references/operation-contract.md file contains a specific 'Operation contract' that prohibits treating results as instructions.
Capability inventory: Write operations like create_database and install_mysql are available but restricted to explicit user authorization.
Sanitization: The skill requires redaction of secrets, API keys, and raw credentials from all exported reports.
[DATA_EXFILTRATION]: The skill includes explicit security guidelines to prevent the leaking of sensitive information. It mandates that API keys and database passwords must never be printed and that sensitive inventory must be redacted from outputs.
[EXTERNAL_DOWNLOADS]: The skill references tool definitions from the official 1Panel GitHub repository. These references are used for documentation and capability scoping rather than runtime code execution from unknown sources.

1panel-security-audit