The Agent Skills Directory

[SAFE]: The skill implements a secure-by-design architecture for handling media generation requests. It enforces explicit idempotency, job state transitions, and cost limits, which prevent resource exhaustion and ensure predictable execution. Furthermore, the instructions explicitly forbid the exposure of provider credentials in logs or diagnostics and require media verification before reporting success.- [PROMPT_INJECTION]: The skill processes external media generation request packages, which represents a potential surface for indirect prompt injection via embedded instructions in generation payloads. However, the skill contains strong mitigation instructions that prohibit the agent from redesigning prompts or production decisions and from rewriting the semantic payload, effectively treating the untrusted data as a pass-through.
Ingestion points: Input payload containing generation_package and provider_routes as defined in schemas/input.schema.json.
Boundary markers: Clear instructions stating 'Do not redesign prompts or production decisions' and 'Submit requests without rewriting their semantic payload'.
Capability inventory: The skill is capable of network communication with providers and writing to durable project storage.
Sanitization: Incoming requests are validated against a JSON schema and require an upstream quality gate audit before execution.

execute-media-generation