resilience-flow-control
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is primarily composed of markdown-based documentation and reference papers for technical systems. It does not contain any executable scripts, binaries, or commands that perform network or file system operations.
- [INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface for indirect prompt injection by processing user-supplied intent to guide code modifications.
- Ingestion points: User input is ingested via the 'Select the Operating Mode' section in
SKILL.mdto trigger Think, Review, Change, or Verify workflows. - Boundary markers: The skill uses a 'Domain Law' and a 'Stop Conditions' section to provide structural boundaries and ensure adherence to safety principles.
- Capability inventory: The 'Change' mode involves modifying repository contents based on user-driven requirements, which is an exploitable capability.
- Sanitization: The workflow requires the agent to validate all decisions against normative engineering requirements and perform failure-injection testing, which serves as a verification layer.
Audit Metadata