The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements robust practices for secret management. It explicitly forbids printing the BUZZ_PRIVATE_KEY in logs, command arguments, or terminal output. It correctly identifies that sensitive information should be stored in environment files (.env) and passed to subprocesses via the environment rather than CLI flags.
[COMMAND_EXECUTION]: The skill uses local CLI tools (hermes, buzz, ffprobe) to perform tasks. All state-changing operations, particularly in the self-hosting workflow, are designed to be user-approved rather than fully autonomous, providing a safety gate against unintended system changes.
[INDIRECT_PROMPT_INJECTION]: The skill integrates with external messaging sources, creating a potential surface for indirect injection.
Ingestion points: Inbound events and messages from the Buzz relay (referenced in references/setup.md).
Boundary markers: No specific delimiters are defined for untrusted message content, but the skill relies on evidence-based reporting stages.
Capability inventory: Subprocess execution of the buzz CLI and hermes gateway, along with file reads for media attachments (references/media.md).
Sanitization: Instructions strictly require redacting private keys and sensitive identifiers from logs, shell history, and rendered output.

omh-buzz