The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of the griffin CLI, which is a vendor-owned resource for 'griffin-open-source'. The instructions specify the use of the --json global flag to ensure the agent receives structured, machine-readable data, which is a key safety measure for command execution.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface as it processes monitor files (JavaScript and TypeScript) within the project.
Ingestion points: Monitor files located in __griffin__ directories, discovered via the griffin validate or griffin test commands.
Boundary markers: The skill enforces an execution contract requiring the agent to parse exactly one JSON object from stdout/stderr, which serves as a delimiter to prevent the agent from treating untrusted monitor output as valid instructions.
Capability inventory: Subprocess calls to the griffin binary, including capabilities to modify cloud resources (apply, destroy) and manage sensitive configuration (secrets set).
Sanitization: The skill relies on the CLI's internal validation (griffin validate) and the --json output format to ensure data integrity during processing.
[CREDENTIALS_UNSAFE]: The skill manages authentication tokens and secrets. It uses a standard vendor-specific path (~/.griffin/credentials.json) for credential storage and utilizes a two-step device flow for browser-based authentication, ensuring the agent does not directly handle the user's primary login credentials.

griffin-cli