icp-research

Pass

Audited by Gen Agent Trust Hub on Apr 1, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [SAFE]: No malicious patterns, obfuscation, or unauthorized data access were found. The skill maintains operational safety through structured agent interactions and a quality control agent.
  • [COMMAND_EXECUTION]: The script scripts/search-platforms.sh is a benign helper utility used to generate search queries for platforms like Reddit and LinkedIn. It does not execute commands or network operations.
  • [PROMPT_INJECTION]: The skill processes untrusted research data (customer quotes), but the risk is mitigated by its ingestion architecture: 1. Ingestion points: agents/voc-collector-agent.md. 2. Boundary markers: Structured dispatch prompts in SKILL.md. 3. Capability inventory: Generates markdown artifacts and performs network operations via research tools. 4. Sanitization: agents/critic-agent.md quality gates.
  • [DATA_EXFILTRATION]: The skill reads project context from .agents/product-context.md. No unauthorized external communication or credential harvesting patterns were detected.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 1, 2026, 06:48 PM