The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to clone an external GitHub repository (NousResearch/pokemon-agent) and perform a package installation via pip install in editable mode, which allows for the execution of arbitrary code from a third-party source.- [DATA_EXFILTRATION]: The instructions direct the agent to establish an SSH reverse tunnel (ssh -R 80:localhost:9876 nokey@localhost.run) to the localhost.run service. This action exposes a local network port to the public internet, creating a potential vector for unauthorized external access or exfiltration of sensitive session data.- [COMMAND_EXECUTION]: The skill utilizes multiple shell commands including git clone, python3 -m venv, pip install, and background service management commands to initialize and run the emulator environment.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from an external source (emulated game RAM and screenshots) to drive agent decision-making through vision models, creating an attack surface for data-driven influence. Ingestion points: GET /state and GET /screenshot endpoints defined in SKILL.md. Boundary markers: None present. Capability inventory: subprocess execution (ssh, pip, git) and local file writing. Sanitization: No sanitization or verification of external game data or vision model outputs is specified.

pokemon-player