The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the firedrill CLI tool to initialize, build, and interact with synthetic services. These commands (e.g., firedrill tools publish, firedrill tools seed, firedrill tools invoke) are standard for the intended workflow.
[DYNAMIC_EXECUTION]: The skill's primary function is the authoring of TypeScript/JavaScript code (service.ts) that is subsequently published and executed as a synthetic service on a remote host. The documentation notes that these handlers are executed in a restricted environment with limited capabilities (e.g., no filesystem, no network access, and isolated state).
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external data through firedrill tools seed and service logic via firedrill tools publish. This creates a potential surface for indirect prompt injection if the authored content contains malicious instructions intended to influence the agent.
Ingestion points: Service manifest (manifest.json), source code (service.ts), and dataset files (seed.json) in the references/authoring.md file.
Boundary markers: The documentation in SKILL.md and references/authoring.md includes clear warnings to use explicit upload lists and sanitized seed records, and to never upload live customer data.
Capability inventory: The agent can publish code for remote execution and invoke operations that interact with that code's state, as seen in references/authoring.md and references/services.md.
Sanitization: The skill instructions explicitly direct the user to remove customer secrets and sanitize records before upload, although no automated sanitization logic is provided within the skill instructions.

firedrill-service-authoring