The Agent Skills Directory

[SAFE]: The skill implements a structured framework focusing on traceability, cognitive integrity, and testing protocols. These instructions steer the agent toward deterministic behavior.
[INDIRECT_PROMPT_INJECTION]: The skill identifies external web content (e.g., via BrowserOS) as a potentially hostile input source. It provides clear instructions to treat such data as untrusted and to redact PII or secrets before sharing context with subagents, which serves as a defensive measure for this attack surface.
[COMMAND_EXECUTION]: The instructions mention the use of an ExecutionProxy Interface (using the rtk prefix) for running commands like npm test. This describes a structured and transparent way for the agent to interact with the environment for testing purposes within the defined framework.
[DATA_EXPOSURE_EXFILTRATION]: Explicit instructions are provided to protect sensitive data and log operations into a persistent MemoryStore (e.g., Obsidian), adhering to the principle of least privilege.

audit-website