Security audit
Security checks for vulnerabilities and agentic risk
This is an instruction-only shopping consultation skill for choosing power bank specs, with no code execution, credentials, persistence, or hidden data handling.
Before installing, understand that the skill may ask for device models, travel context, and country or region, then may suggest real power banks as research starting points. Verify current prices, availability, certifications, and airline rules before buying.
Line L039 says the agent will not recommend specific brands and that the goal is to let the user evaluate products independently. The very next instruction at L040 contradicts this by saying the agent will suggest a small number of real power banks, creating a direct intent-code/documentation divergence within the skill definition.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
**User** says "I want to charge my phone and laptop" without specifying models.
**Agent action:** "To calculate the capacity you need, I need to know the battery size of your phone and laptop. Could you tell me the make and model? For example, an iPhone 15 has a roughly 3,274 mAh battery, while a MacBook Air M2 has a 52.6 Wh battery."
**Agent does NOT:** Assume generic values and proceed without asking.
### Example 3 — User skips to brands
The manifest description frames the skill as a brand-neutral advisor focused on calculating required capacity, wattage, and protocol. However, the documented behavior later requires suggesting up to five specific commercial products, which goes beyond purely brand-neutral specification guidance.