toolify

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). SKILL.md Step 2 (“Fetch official setup docs”) uses WebFetch or context7:query-docs to pull current official quickstart at runtime, which can involve fetching public web content (outsider-authored text) and then reading it into the agent context.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly performs a runtime fetch of the live official quickstart using WebFetch / context7 (WebFetch ), and that fetched documentation is read and used to drive scaffolding and pin SDK versions, so external content directly controls the generated prompts/instructions ().

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill explicitly targets payment gateway integrations (names Stripe, LemonSqueezy, Paddle), includes payment-specific scaffolding (webhook signature verification, customer model, event handlers), and provides payment smoke-test examples using a Stripe secret key. It also references a Stripe-specific recipe. These are concrete, payment-focused integrations (payment gateway APIs) that would enable an agent to execute or wire up financial operations.