incident
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a workflow surface where external data (e.g., from BrowserOS) is ingested. It provides defensive guidance to mitigate potential injection risks.
- Ingestion points: Processes external data and web content via the BrowserOS interface (SKILL.md).
- Boundary markers: Instructions are included to redact secrets and PII before sharing context with subagents (SKILL.md).
- Capability inventory: The skill utilizes an ExecutionProxy for terminal commands and a MemoryStore for writing persistent markdown artifacts (SKILL.md).
- Sanitization: Explicitly mandates that external web content and data be treated as hostile (SKILL.md).
- [SAFE]: The 'INCIDENT Mode' described in the skill allows for bypassing standard planning phases (such as PRD and Architecture blueprints) during emergency hotfixes. This is a legitimate workflow optimization and does not attempt to bypass security guardrails. The skill reinforces security through deterministic execution and strict context validation protocols.
Audit Metadata