The almost-right build.
The agent built most of it. But “most” isn’t a list you can review.
Make every agreement explicit THE MISSING STEP BETWEEN BUILD AND SHIP
Your agent built it.
Is it what you meant?
A working app isn’t always the app you asked for.
Walkdown brings the spec, the design, and the running build
together—so nothing gets lost on the way to done.
+ THE APP, MEET THE AGREEMENT.01 / FIELD VIEW
Actual app. Actual rule. One place to review.Click to look closer
Your tools.
A shared understanding.
test suite required.
A shared understanding.
Claude CodeFigma exportsPlaywrightRSpecnode:test
No replacementtest suite required.
01 — THE GAPINTENT ≠ IMPLEMENTATION
“Done” is not
a specification.
You have the prototype. You have the pull request. Somewhere between them, the empty state disappeared, the interaction changed, and a decision got buried in a chat.
Your tests can pass while the product still misses the point. Walkdown makes that gap visible.
The moving target.
The requirement changed. Yesterday’s green check didn’t get the memo.
Know when evidence goes staleThe feedback black hole.
“That button, on that screen.” Context lost. Another round of explaining.
Keep the conversation attached02 — ON THE GROUNDLOOK. COMPARE. CLOSE THE LOOP.
Don’t just read the spec.
Walk through it.
The review happens where the product happens.
Your actual app, with the agreement right beside it.
A checklist with receipts.
Not a manually updated status. Every rule shows its checks, agent review, and human signatures—derived from the evidence. Unjudged work stays visible.
REAL CAPTURE / RULE BOARD
BuildDesign
Same screen. No mental gymnastics.
Walkdown fades the prototype over your running app at the same viewport. Try this screenshot comparison to spot what changed—without switching windows.
INTERACTIVE / SCREENSHOT COMPARISONA conversation that becomes a decision.
Pin feedback to an element. Ask a question against a rule. Answers get folded into the blueprint, so the next agent gets the decision—not just another chat to search.
REAL CAPTURE / INCORPORATED QUESTION03 — THE STANDARDEVIDENCE BEFORE CONFIDENCE
AI does the legwork.
You have the final word.
Three kinds of verification. Three different jobs.
Each one earns the next one’s attention.
01 / CHECKS
Prove the behavior.
Your existing tests catch regressions. Tag them to rules; Walkdown connects the results. No second test suite to maintain.
✓ Your test suite, linked.
02 / AGENT WALKDOWN
Question the experience.
An agent reads the rule and tries the app. It compares, investigates, and records evidence before asking for your attention.
✓ A fresh pair of eyes.
03 / HUMAN ACCEPTANCE
Decide if it’s right.
You review what matters and sign for your role. An agent can claim the work is fixed. It cannot accept it on your behalf.
Looks right.SIGNED BY YOU ↗
When the agreement changes, the evidence should too.
A pass against an older rule is marked stale—not quietly carried forward as green.
04 — YOUR WORKFLOW, CONNECTED
A shared blueprint.
Not another silo.
Your designs and requirements become plain-language rules. Your agents build against them. Your team reviews the same agreement in the browser.
Underneath it all? Files you own, a CLI, and an append-only record of what was actually verified.
Explore the thinking behind WalkdownDESIGNREQUIREMENTS
blueprint / invites.yml YAML
id: invites.batch.confirm-count
statement: >
Inviting a batch asks for
confirmation naming how many
people will be emailed.
verify:
- checks
- agent
signoff: [eng, product]YOUR TESTSYOUR AGENTYOU
A FEW THINGS WORTH KNOWING
Before you
walk in.
Is this another testing framework?
No. Walkdown links acceptance rules to the tests you already own. Playwright, RSpec, and node:test have reporters; the runner contract lets other frameworks participate. Your suite stays your suite.
Does my product data leave my machine?
Walkdown’s server runs locally. Blueprints, threads, and verification records are files you control. Any AI agent you choose to use has its own data handling and service terms.
Do I need a finished app to start?
No. You can review and approve the specification before the build exists. Walkdown distinguishes approval of the wording from verification of the working product.
How is this different from Highball?
Highball enforces your repository’s engineering checks while an agent builds. Walkdown asks whether the built product matches the agreed intent. They complement each other; they don’t share a verdict.
Meet Highball ↗ THE LAST STEP IS A HUMAN ONE.
Build with speed.
Ship with understanding.
Give your agent a blueprint. Give yourself the final word.
PASTE THIS INTO YOUR CODING AGENT
Read https://github.com/profoundry-us/walkdown/blob/main/site/setup.md and set Walkdown up for this project.
I’ll set it up myself Open source. MIT licensed. No account.