Security audit
Security checks for vulnerabilities and agentic risk
This is a local, instruction-only retail operations skill that stores optional store notes after confirmation and does not show hidden network, credential, or destructive behavior.
Before installing, be aware that the skill may keep store-operation notes in ~/store/ if you confirm file creation. Do not put payroll records, payment card data, private customer details, or sensitive employee records into those notes, and narrow the activation preference if you only want help when explicitly requested.
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
# Merchandising and Promotions
## Display Rules
- Lead with hero products, clear prices, and one dominant message per display.
- Keep entry and queue zones simple; clutter reduces conversion.
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.
- Never run overlapping messages in the same zone.
- Never keep a promotion live because "it has always been there."
- Never judge a display by personal taste alone; use sell-through and conversion.
- Never put fragile or high-shrink promo items where control disappears.
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.
## Hard Rules
- Never trust the system blindly when the shelf says otherwise.
- Never place a new order without checking existing back stock and inbound deliveries.
- Never mix damaged, returnable, and saleable stock in the same area.
- Never wait until month end to investigate unexplained shrink.
The activation scope is broad enough to trigger on many ordinary retail-adjacent conversations, which can cause the skill to engage without sufficiently clear user intent. In an agent system, this increases the chance of unsolicited guidance, context capture, or workflow steering in situations where the user did not explicitly ask for store-operation support.
Telling the agent to 'start naturally' and help immediately without explicit opt-in can bypass user-controlled activation expectations. Even though the instruction is operational rather than overtly malicious, it encourages implicit activation and reduces transparency about when the skill is influencing the conversation.
No suspicious patterns detected.