skills/quantipixels/skills/pare/Gen Agent Trust Hub

pare

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as source code and configuration files, creating an indirect prompt injection surface. This risk is mitigated by the skill's mandatory 'read-only' constraint and its explicit prohibition of code execution, builds, or environment mutations. * Ingestion points: Audit of software source code, entry-points, and runtime configurations as defined in SKILL.md. * Boundary markers: No specific delimiters or 'ignore' instructions are provided to separate audited data from the agent instructions. * Capability inventory: The agent is limited to file inspection; the skill explicitly disallows subprocess execution, network operations, or file-write capabilities. * Sanitization: No sanitization or escaping of ingested source code content is specified.
  • [SAFE]: No malicious patterns, unauthorized downloads, or exfiltration attempts were detected. The skill's functionality is strictly limited to static analysis and simplification recommendations.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 01:24 PM
Security Audit — agent-trust-hub — pare