The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the pokemon-agent package from the NousResearch/pokemon-agent GitHub repository. NousResearch is a well-known organization in the AI research community.
[COMMAND_EXECUTION]: The skill uses local shell commands to manage the game environment, including starting the pokemon-agent server and establishing an SSH reverse tunnel via localhost.run. The use of localhost.run is a standard method for exposing a local port to a public URL for viewing the dashboard. The instructions also specify writing temporary files to /tmp/pokemon.png for vision analysis.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it processes external game data.
Ingestion points: The agent reads game state via GET /state and processes screenshots of the emulator via GET /screenshot (SKILL.md).
Boundary markers: There are no specific boundary markers or instructions to ignore data embedded within the game visuals or RAM state.
Capability inventory: The skill can execute CLI tools (ssh, pokemon-agent), write files to the local filesystem (/tmp/pokemon.png), and interact with local network services (SKILL.md).
Sanitization: No explicit sanitization or filtering is performed on the game state data before it is processed by the vision model.

pokemon-player