The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for utilizing the relaymessenger CLI for agent management and includes code for spawning ffmpeg subprocesses via asyncio.create_subprocess_exec to handle video tracks in calls. The command usage is restricted to the skill's documented functionality and follows established development patterns.
[EXTERNAL_DOWNLOADS]: The instructions reference installing the relaymessenger package and the @relaymessenger/sdk from official registries. It also utilizes well-known libraries such as pipecat-ai for voice provider integration.
[INDIRECT_PROMPT_INJECTION]: The skill consumes untrusted data through incoming message events (message.received) and user reviews (rating.created).
Ingestion points: references/agent-events.md and references/build-an-agent.md describe handling inbound WebSocket events.
Boundary markers: The skill explicitly instructs the agent to treat all incoming labels, values, and review text as untrusted data rather than instructions.
Capability inventory: The provided code references have capabilities for file system writes (storing chat IDs), network operations (Relay API), and subprocess execution (ffmpeg).
Sanitization: The instructions emphasize that the agent must treat external content as data and never as commands.

relay