endpoint-definition-designer
Pass
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill instructions do not contain any patterns typical of prompt injection or instructions to bypass safety guardrails. It includes explicit 'Out-of-scope' redirects to prevent the agent from providing medical advice or patient-specific predictions.
- [DATA_EXFILTRATION]: No network operations or sensitive file access patterns were detected. The skill does not request access to credentials, environment variables, or local configuration files. A single reference to the author's public GitHub repository is provided as a source, which is a standard and safe practice.
- [REMOTE_CODE_EXECUTION]: There are no remote code execution patterns. The skill does not use package managers, download remote scripts, or invoke shell commands. The inclusion of an evaluation report (eval_report_endpoint-definition-designer_result.json) confirms that the skill is intended for design planning rather than code generation.
- [OBFUSCATION]: All analyzed files, including the SKILL.md and reference markdown files, are written in plain, human-readable text. No Base64, hex encoding, zero-width characters, or hidden steganographic patterns were identified.
- [INDIRECT_PROMPT_INJECTION]: While the skill ingests untrusted user data (e.g., study objectives), it lacks any exploitable capabilities such as tool execution, network access, or file-writing. The risk is negligible as the agent only generates structured text based on the input.
Audit Metadata