Testing vs Design · both proven

differential-review vs prototype

Both skills were run through Claude twice: once on a fixed brief without the skill, once with it. The briefs differ because each is written for the job the skill claims to do, so compare each skill against its own baseline, then decide which difference matters for your work. They are in different categories, so this is a choice of job, not a head-to-head on the same job.

differential-review

Performs security-focused differential review of code changes. Adapts analysis depth to codebase size, uses git blame for context, calculates blast radius by counting callers, checks test coverage of modified code, and generates a markdown report. Use when reviewing a PR, commit, or diff for security vulnerabilities, checking whether a change re-introduces a previously fixed bug, asking what else…

Category: Testing
Source: trailofbits/skills
License: Other
Proof model: claude-opus-5-5
Brief: “Do a security-focused review of this change and present it as a single HTML page. Commit: "Allow admins to impersonate users for support" +app.get('/admin/impersonate/:id', async (req, res) => { + if (req.cookies.role !== 'admin') return res.sendStatus(403); + const user = await User.findById(req.params.id); + req.session.userId = user.id; + res.redirect(req.query.next || '/'); +});”
Skill commit: 2be6433

Without the skillWith differential-review

⇆

Without, runWith differential-review, run3 independent runs per side, none selected by hand.

Download the with-skill page

prototype

Build a throwaway prototype to answer a design question. Use when the user wants to sanity-check whether a state model or logic feels right, or explore what a UI should look like.

Category: Design
Source: mattpocock/skills
License: MIT
Proof model: claude-opus-5-5
Brief: “We are designing the order state model for a food delivery app. An order can be placed, accepted by the restaurant, being prepared, ready for pickup, picked up by a courier, and delivered. The customer can cancel until the restaurant starts preparing; the restaurant can cancel at any time before pickup; a cancellation after payment triggers a refund, partial if food was already prepared. We are not sure this logic is right. Build a throwaway prototype as a single HTML page that lets someone on the team drive an order through these states and see where the rules break or get confusing.”
Skill commit: 5beb35b

Without the skillWith prototype

⇆

Without, runWith prototype, run3 independent runs per side, none selected by hand.

Download the with-skill page

We don’t pick a winner for you: the pages are the evidence. Both proofs can be repeated with the brief, model and commit shown above. See alternatives to differential-review or alternatives to prototype.