agent-md-tuner

Pass

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill uses a limited set of tools (Read, Write, Edit, Glob, Grep) and does not request high-risk capabilities such as Shell execution or Network access.
  • [SAFE]: No evidence of data exfiltration or credential harvesting was found. The skill processes project configuration files locally to provide recommendations.
  • [SAFE]: No obfuscation techniques, hidden commands, or malicious encodings (e.g., Base64, homoglyphs) were detected in any of the skill's files.
  • [SAFE]: The skill includes a clear 'Human-in-the-loop' requirement, explicitly instructing the agent to show proposed changes and wait for user approval before writing to the disk.
  • [SAFE]: References to external sources (e.g., the GitHub repository for Karpathy Principles) are provided for informational purposes and originate from well-known services.
  • [SAFE]: An indirect prompt injection surface is present as the skill reads untrusted project files (Phase 1 and 2), but the operational scope is limited to auditing and templating, and there are no high-privilege side effects like network exfiltration or command execution associated with the data ingestion.
Audit Metadata
Risk Level
SAFE
Analyzed
May 20, 2026, 08:37 AM
Security Audit — agent-trust-hub — agent-md-tuner