product-taste-intuition
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFENO_CODE
Full Analysis
- Prompt Injection (SAFE): No patterns were detected that attempt to bypass safety guardrails, override agent instructions, or reveal system prompts. All instructions are focused on the primary task of product critique.
- Data Exposure & Exfiltration (SAFE): The skill does not contain code to access sensitive files (e.g., SSH keys, environment variables) or exfiltrate data. While the workflow mentions analyzing product data (replays, funnels), it does not provide scripts or credentials to access these sources.
- Unverifiable Dependencies & Remote Code Execution (SAFE): There are no scripts, package files (e.g., package.json, requirements.txt), or remote download commands (e.g., curl | bash) included in the skill.
- Indirect Prompt Injection (SAFE): The skill involves studying external 'benchmark' products, which creates a surface for indirect prompt injection if those products contain malicious hidden text.
- Ingestion points: External benchmark descriptions and product behavior observed during study sessions (Workflow Step 4).
- Boundary markers: The skill uses rigid Markdown templates (
TEMPLATES.md) to isolate external observations from the agent's logic. - Capability inventory: Local file-writing capability for saving the Calibration Pack.
- Sanitization: None explicitly defined in the skill logic, though the structured workflow serves as a functional delimiter.
- Obfuscation (SAFE): No Base64 encoding, zero-width characters, or homoglyphs were detected in the text or metadata.
- Privilege Escalation & Persistence (SAFE): No commands related to administrative access or system persistence are present.
Audit Metadata