engineering-harness

Pass

Audited by Gen Agent Trust Hub on May 4, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • [SAFE]: The skill is designed strictly for analysis and planning. It explicitly forbids the agent from implementing code changes, which ensures that it remains within a safe operational boundary.\n- [NO_CODE]: The extension consists only of Markdown files and configuration. The absence of scripts (Python, Node.js, etc.) removes risks of malicious code execution or unintended side effects.\n- [PROMPT_INJECTION]: The skill ingests untrusted repository data and user-provided harnesses, presenting a surface for indirect prompt injection. This is evaluated as safe because the skill does not possess tools for network exfiltration or command execution.\n
  • Ingestion points: Repository source files and documentation (SKILL.md, references/context-curation.md).\n
  • Boundary markers: None defined in the template instructions.\n
  • Capability inventory: Local file reading and summarization.\n
  • Sanitization: Not explicitly implemented in the skill instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
May 4, 2026, 08:33 AM