scientific-analysis-architect
Pass
Audited by Gen Agent Trust Hub on Feb 28, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill follows a structured phase-based architecture with explicit orchestration rules and user approval gates. No malicious patterns or obfuscation techniques were detected.
- [SAFE]: Regarding indirect prompt injection (Category 8), the ingestion points are located in Phase 1 and Phase 2 (SKILL.md) where user-provided dataset and research descriptions enter the context. Explicit boundary markers and sanitization for this external input are absent. The skill's capability inventory includes file writing (Write tool in Phase 0, 5, and 7), bash-based directory validation (Bash tool in Phase 0), and multi-agent spawning (Task tool in Phase 2, 4, and 5). However, the skill primarily generates markdown-formatted pseudocode rather than executing runtime code, and its structured logic prevents the escalation of untrusted data into critical system operations.
- [SAFE]: The skill does not perform external downloads or install third-party packages. It relies on standard internal tools and follows an offline-first session management strategy.
- [SAFE]: File system access is restricted to session directories and user-specified output paths. Access to internal ~/.claude/ reference documents is used strictly for workflow compliance and does not involve sensitive credential exposure or data exfiltration.
Audit Metadata