computational-experiments

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill is designed to scaffold, modify, and execute Python code to run computational experiments. It includes an 'Autonomous Sweep' and an 'Explore Loop' that modify code based on metrics and rerun it using uv run python. This behavior is the primary purpose of the skill for algorithmic research.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted project context files.
  • Ingestion points: Reads CLAUDE.md, MEMORY.md, and .context/project-recap.md in Phase 1 to detect project structure and conventions.
  • Boundary markers: None mentioned in the instructions for these files.
  • Capability inventory: Includes Bash, Write, Edit, and Glob tools across all phases to manage files and execute research code.
  • Sanitization: No specific sanitization or filtering is described for the content of these context files before they influence the agent's scaffolding or coding decisions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 06:16 AM
Security Audit — agent-trust-hub — computational-experiments