ghost-line
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external replay data and IDs from potentially untrusted sources, creating an injection surface. Although the skill mandates isolation from live simulation parameters, it lacks explicit prompt-level delimiters or sanitization for ingested content.\n
- Ingestion points:
ghost_replay_idand external ghost sources (SKILL.md).\n - Boundary markers: The instructions focus on schema validation and compatibility predicates rather than explicit boundary delimiters.\n
- Capability inventory: Interacts with
gameplay-harnessfor deterministic tape infrastructure (SKILL.md).\n - Sanitization: Employs identity and integrity records to validate downloaded data and distinguish defects from stale data (
SKILL.md).
Audit Metadata