The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill explicitly acknowledges the ingestion of untrusted data from external sources, creating a potential surface for indirect prompt injection attacks.
Ingestion points: Web content and external data accessed via tools like BrowserOS (mentioned in the Security & Multi-Agent Hygiene section).
Boundary markers: The instructions do not define specific delimiters or boundary markers for untrusted content, though they provide a high-level protocol to treat such data as hostile.
Capability inventory: The skill utilizes the ExecutionProxy Interface (e.g., rtk npm test) which allows for terminal command execution across various scripts and development tasks.
Sanitization: The protocol mandates the redaction of secrets and PII (Personally Identifiable Information) before sharing context with subagents, acting as a manual sanitization step.
[COMMAND_EXECUTION]: The skill defines an execution model where all terminal actions are routed through an ExecutionProxy Interface (using the rtk prefix). While intended for auditing and efficiency (Token Economy), this provides the agent with a mechanism to execute shell commands (e.g., npm test) on the host environment.

marketing