human-gate-designer
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a design pattern for workflows that ingest and process human feedback, which introduces an attack surface for indirect prompt injection in the target implementation.
- Ingestion points: The 'Feedback Injection' section in
SKILL.mddescribes collecting 'text input for human feedback' via a UI gate. - Boundary markers: The recommended prompt template uses basic labels (
Human feedback: "[...]") but does not explicitly instruct the agent to treat the content as data or provide strict delimiters to prevent command injection from the human reviewer. - Capability inventory: The skill itself uses standard
Read,Write, andEdittools; however, the gates it designs are intended to trigger re-execution of other nodes in a DAG. - Sanitization: The skill does not currently provide guidance on sanitizing or escaping the human-provided feedback before it is interpolated into the re-execution prompt.
Audit Metadata