Release gate for AI-generated work

Check AI patches before merge.

Send a generated patch with its task, requirements, and evidence. Get a release verdict and specific fixes before review. The same API checks code, UI, documents, images, data, and mixed packages.

Check a patch free Add the GitHub check Connect an agent Get access — $249

Send the work

Include the task, requirements, and evidence.

Get a decision

Pass, fail, or insufficient evidence.

Fix what matters

Turn each finding into a focused repair.

Check it again

Confirm the repair with fresh evidence.

7 formatsfrom code and documents to mixed packages

3 verdictspass, fail, or insufficient evidence

Sync + asyncrun now or collect the result later

Repair historysee what changed and what was verified

How it works

Judge the work against the job.

Give it the brief

Send the completed work with the task, requirements, and any evidence needed to check it.

Get a usable result

Each finding names what was observed, what was expected, how to fix it, and how to verify the change.

Keep uncertainty visible

If the submitted evidence cannot support a decision, TUC says so instead of guessing.

Close the repair loop

Submit the revision, run the check again, and keep the full repair history.

GitHub Action

Put the check on every pull request.

.github/workflows/tuc.yml

name: TUC patch gate
on: [pull_request]
permissions:
  contents: read
jobs:
  tuc:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: marksignals/the-ultimate-critic@v1
        with:
          task: Review pull request — ${{ github.event.pull_request.title }}

Default behavior

pass                  merge check passes
fail                  merge check blocks
insufficient_evidence merge check blocks

public repository     enabled
private repository    blocked by default

View setup and inputs on GitHub. The free check accepts a diff up to 10,000 characters and does not save the submitted patch or result.

API

A result your workflow can use.

Python SDK request

result = critic.evaluate({
  "artifact_type": "code",
  "artifact": {
    "content": generated_code,
    "metadata": {
      "builder_model": builder_model
    }
  },
  "intent": {
    "prompt": task,
    "requirements": requirements,
    "constraints": constraints
  },
  "domain": "software",
  "evaluation_profile": "code_review"
})

Handle the verdict

verdict = result["overall"]["verdict"]

if verdict == "insufficient_evidence":
    capture_more_evidence()
elif verdict == "fail":
    tickets = create_repairs(result["findings"])
    send_to_repair_agent(tickets)
else:
    release(result["evaluation_id"])

# Your workflow decides what happens next.

What it checks

One API for the work your system makes.

Text

Code

Image

JSON

Game scene

Mixed package

Founding access

Use it on one live workflow.

Founding access

A 90-day pilot to connect one real pipeline and decide whether TUC earns a permanent place.

Up to two hours of first-pipeline setup
1,000 review credits
Up to 1,000 standard checks
Up to 100 full mixed-package reviews
Code, UI, documents, images, and data
Pass, fail, or insufficient evidence
Repair verification
One tenant for 90 days

Buy founding access — $249

Not sure it fits? Run the free preview or send your workflow details. The fit form accepts no payment or work samples.

See TUC run the repair loop on a visual identity or sample code. Read the terms or contact us.

$249

one-time · not a subscription

Standard checks use 1 credit, extended packages use 5, and full mixed-package reviews use 10. Quote the charge before the agent submits.

Questions

What can TUC check?

Code, interfaces, documents, images, structured data, game scenes, and mixed packages.

What comes back?

A verdict, score, findings, recommended fixes, verification steps, and a report.

How do credits work?

Standard checks use 1 credit, extended packages use 5, and full mixed-package reviews use 10. Rejected requests use no credits.

What if evidence is missing?

The verdict is insufficient_evidence. Your workflow can collect more proof and try again.

Is this a subscription?

No. Founding access is a one-time $249 purchase for 90 days and 1,000 review credits.