The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions specify pulling data (canvas, reports, diagrams) using a "Chrome session token". Accessing or utilizing active browser session tokens is a sensitive operation as it grants the agent authenticated access to the user's web session.
[DATA_EXFILTRATION]: The skill performs network operations via GET /api/updates?company=<slug> and POST /api/update. These endpoints are used to synchronize findings across agents, which involves transmitting data derived from internal transcripts and process concepts to a network service.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, creating a surface for indirect injection attacks.
Ingestion points: The agent reads transcript passages, canvas data, and external information retrieved via WebSearch and WebFetch during its critique phase.
Boundary markers: The instructions do not define specific delimiters or warnings to prevent the agent from following instructions that might be embedded within the transcripts or web content.
Capability inventory: The skill possesses network capabilities (API POST/GET) and web browsing tools (WebSearch, WebFetch) to interact with external environments.
Sanitization: There are no mentions of sanitizing, filtering, or validating the content retrieved from transcripts or the web before it is processed by the agent.

processflow-critic