reconsider
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from the user and external sources, forming a standard ingestion surface for indirect prompt injection which is managed through defensive logic.\n
- Ingestion points: Utilizes user-provided facts, conversation memory, local files, and external authoritative sources as defined in the 'Evidence Ladder' (SKILL.md).\n
- Boundary markers: Specifically recommends isolating subagents from the main conversation history and providing only the minimum necessary context via 'task packets' to prevent instruction bleed.\n
- Capability inventory: Mentions tool capabilities including subagent delegation (
agent-team) and isolated execution (codex exec) for verification tasks.\n - Sanitization: Implements logical sanitization layers, including 'frame challenging', independent scrutiny, and failure-mode analysis to identify and mitigate unreliable or malicious context.
Audit Metadata