agent-harness-engineering
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a high-capability agent harness that processes untrusted external data, creating a surface for indirect prompt injection attacks.
- Ingestion points: Data is ingested from web sources, sub-agent summaries, and project-local files.
- Boundary markers: The charter mandates structural marking of untrusted content (e.g., pre-pending warnings) and explicit delimiters for external inputs.
- Capability inventory: The system possesses tools for file system access, shell execution, and network egress, which are constrained by mandatory sandboxing and human-in-the-loop gates.
- Sanitization: Prescribed mitigations include telemetry sanitization, 2-stage approval classifiers, and an 'Intersection' authorization model (Agent baseline ∩ caller role ∩ host-level permissions) to enforce least privilege.
Audit Metadata