agent-starters
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a framework where agents ingest 'startup context' from external YAML registry files. This ingestion process creates a vulnerability surface for indirect prompt injection if the registry sources are compromised.
- Ingestion points: Registry manifest files located in the
<registry-root>or the local directory~/.openrig/agent-starters. - Boundary markers: The instructions mention a
guidance_mergestartup file that acts as a delimiter for the delivered guidance. - Capability inventory: The system performs registry resolution, seat expansion (
rig expand), and transitions through a six-state lifecycle (captured, named, inspectable, used, promoted, deprecated). - Sanitization: The skill provides defensive guidelines for the resolver, including checking 'entry shape' and enforcing a 'credential boundary' to prevent the leakage of authentication material.
Audit Metadata