The Agent Skills Directory

[SAFE]: The skill is a standard technical manual for self-hosting a software stack. It does not contain obfuscated code, hidden instructions, or attempts to bypass safety filters.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill identifies sensitive local files like .env and managed-agents.json (which contains private keys) to provide necessary troubleshooting context. It includes a specific security warning to treat these files as sensitive and provides backup commands that stay within the local environment.
[COMMAND_EXECUTION]: The skill provides practical CLI commands for Docker, PostgreSQL, and PowerShell. These are used for legitimate deployment, debugging, and backup tasks. The instructions emphasize that the user should execute state-changing commands themselves.
[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it instructs the agent to analyze log files that could contain untrusted data from the network. This is documented as a vulnerability surface.
Ingestion points: Agent harness logs and relay logs.
Boundary markers: The skill explicitly tells the agent to 'Guide, don't drive' and let the user run commands that change state.
Capability inventory: The skill uses docker compose, bash, powershell, and curl.
Sanitization: No specific sanitization of log content is mentioned; the skill relies on the user to review suggested commands.

buzz-self-hosting