Security audit
Security checks for vulnerabilities and agentic risk
This is a non-executable CRO guidance skill with some broad routing and privacy-caveat gaps, but its behavior is disclosed, purpose-aligned, and proportionate.
Before installing, be aware that the skill may look for local product-marketing context files and may suggest CRO tests involving visitor segmentation, gating, tracking, or field enrichment. Use those recommendations with normal privacy, consent, and compliance review, especially when collecting or personalizing based on user data.
The trigger description is extremely broad and includes many generic phrases like 'this page needs work,' 'low conversion rate,' and even invocation when a user merely shares a URL and asks for feedback. This can cause the CRO skill to activate in situations better handled by other, more specific skills, leading to incorrect routing, irrelevant guidance, and reduced trust in the agent's behavior.
The eval explicitly rewards triggering CRO analysis on the vague phrase 'this page isn't converting,' which broadens activation beyond clear CRO requests and can cause the skill to engage on underspecified inputs. In an agent setting, overly broad routing criteria increase the chance of misclassification, causing the wrong skill to handle requests and potentially bypass more appropriate specialized flows such as signup or onboarding analysis.
This markdown file proposes personalized messaging based on visitor data/source and later suggests dynamic personalization by segment, but it does not mention any privacy, consent, or data-use warning. For markdown files, SQP-2 applies when descriptions omit warnings about behaviors that could affect user data or privacy.
The file suggests gated content and download tracking, which can involve collecting identifiable engagement data, but provides no disclosure about privacy or consent considerations. In markdown guidance, omission of warnings for behaviors affecting privacy is in scope for SQP-2.
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.
## Field-by-Field Optimization
### Email Field
- Single field, no confirmation
- Inline validation
- Typo detection (did you mean gmail.com?)
- Proper mobile keyboard
The skill instructs the agent to proactively read .agents/product-marketing.md or .claude/product-marketing.md before asking questions, which can cause access to project-local context that may contain sensitive business or strategic information without explicit user consent at run time. While the purpose is contextual relevance rather than exfiltration, it expands the agent's data access scope beyond the immediate form-CRO task.
No suspicious patterns detected.