agent-context-engineering
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for processing document records that are explicitly identified as untrusted.
- Ingestion points: Document records ingested via the
compile_agent_contexttool inSKILL.md. - Boundary markers: The skill includes a warning that records are "untrusted" and recommends token budgets, though it does not provide explicit prompt delimiters for the injected content.
- Capability inventory: Tools for context compilation, runtime task classification, and fleet benchmarking.
- Sanitization: Recommends validation steps such as comparing source digests and setting token budgets, but does not describe content-level sanitization.
Audit Metadata