The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an external communication channel, creating a surface for potential indirect prompt injection. The documentation provides mitigation strategies, such as validating message provenance.\n
Ingestion points: The watchdog.mjs and watchdog-exit.mjs scripts feed messages from the WebSocket channel to the agent's stdout.\n
Boundary markers: The skill instructs the agent to differentiate between the 'Provenance' of a message and its 'Body'.\n
Capability inventory: The skill uses the iskron_channel tool and executes local Node.js scripts that use fetch and WebSocket.\n
Sanitization: The skill emphasizes checking message origins and version constraints but does not define specific content filtering.\n- [COMMAND_EXECUTION]: The skill involves the execution of local Node.js scripts (watchdog.mjs, watchdog-exit.mjs) to manage the communication socket. These scripts perform standard networking tasks and are transparent in their operation.\n- [DATA_EXFILTRATION]: The skill manages WebSocket authentication tokens. It demonstrates security awareness by recommending restricted file permissions (0600) and using shell wrappers to hide secrets from the process table (ps).

standing