gtm-icp

Pass

Audited by Gen Agent Trust Hub on Jun 27, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No malicious patterns, prompt injections, or unauthorized data access were identified. The skill's behavior aligns with its described purpose of marketing research and synthesis.
  • [DATA_EXPOSURE]: The skill reads from and writes to the projects/ directory (gtm-context.md and icp-personas.md). This localized file access is necessary for maintaining project context and saving results, and does not involve sensitive system paths or credentials.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external web sources (e.g., G2, Reddit, LinkedIn) via parallel research sub-agents.
  • Ingestion points: External websites via sub-agent research tasks in SKILL.md (Step 2).
  • Boundary markers: The sub-agent prompts use structured instructions to focus on language mining and buying behavior.
  • Capability inventory: Read, Write, WebSearch, and Task tools. The primary capability is writing markdown files to the local project directory.
  • Sanitization: The skill relies on the synthesis step to organize research findings into a structured format before saving.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 27, 2026, 12:52 PM
Security Audit — agent-trust-hub — gtm-icp