Prompt Entropy

Illustrative result: security review prompt

This example shows how Prompt Entropy can turn a vague request into a bounded, checkable instruction. The output below is an example prompt, not a claim that a codebase was inspected.

Revised prompt

Review the provided source code for security vulnerabilities, prioritizing SQL injection.

Scope and authority
- Analyze the accessible code only. Do not modify files or apply fixes.
- Ignore formatting, style, and lint issues.
- Do not infer vulnerabilities without evidence in the supplied code.

For each finding, report:
1. Severity and a concise title.
2. File and line location, or the nearest available code reference.
3. The relevant code path and why it is vulnerable.
4. A concrete remediation that addresses the cause.

If no SQL injection issue is found, state that this specific check found none. Do not claim a complete security audit unless other vulnerability classes were also examined.

What changed

• Kept the requested focus on SQL injection.
• Preserved the instruction to ignore formatting.
• Made the analysis-only boundary explicit.
• Added evidence and location requirements while limiting any “no findings” statement to the checked scope.

Next step

Give the sub-agent access to the intended source files and specify any project-specific database framework or review rules.

Huan Dev skill group

This skill is part of the six-skill Huan Dev collection: https://github.com/HuanHuanHuanFFF/huan-dev