replica-architect
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from
replica/recon.mdandreplica/features.csvwhich are derived from external sources. While this provides an ingestion point for potentially malicious instructions, the skill does not have the tools required to execute code or exfiltrate data.\n - Ingestion points: Reads from
replica/recon.mdandreplica/features.csv(SKILL.md).\n - Boundary markers: None found in instructions.\n
- Capability inventory: Only text writing to documentation and SQL files is permitted. No network or system execution capabilities.\n
- Sanitization: Input data is not sanitized before use.\n- [SAFE]: The stack recommendations and design patterns are standard and follow best practices. No indicators of prompt injection, obfuscation, or credentials exposure were identified.
Audit Metadata