The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/server_bootstrap.py script downloads and executes installation scripts from tailscale.com and beszel.dev. These downloads are used to install official Tailscale and Beszel agent software during the server bootstrap process.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run across scripts/host-governance.py, scripts/server_bootstrap.py, and scripts/jenkins_api.py. These commands are used for system administration tasks such as modifying SSH configurations, managing the UFW firewall, installing packages via apt-get, and interacting with Jenkins APIs.
[INDIRECT_PROMPT_INJECTION]: The skill architecture relies on resolving task contracts and instructions from a 'host infrastructure repository' provided at runtime. An attacker who gains control over the repository's configuration files (.agents/skills-config/host-governance/) could define malicious command templates or parameters that the agent would then execute. This risk is mitigated by the skill's requirement for human authorization (--authorized flag) before executing any non-read-only operations.
[DATA_EXPOSURE]: The skill handles sensitive credentials including Tailscale auth keys, Jenkins API tokens, and SSH bootstrap passwords. The implementation includes logic in scripts/jenkins_api.py to redact secrets from output and instructions in scripts/server_bootstrap.py to ensure credentials are provided via environment variables rather than command-line arguments to prevent leakage in process lists.

host-governance