Open-source · Local-first · Zero runtime dependencies

AI said “done”. Proof checks if it is.

Turn a human task into an acceptance contract, execute real evidence, and get an honest ship decision before production.

Install Proof ↓ ★ GitHub

MIT License Node 18+ 10/10 tests 0 vulnerabilities

proof ship

$ proof ship

TASHEV PROOF

Task: Password reset

reading acceptance contract…

✓ auth regression tests

✓ reset endpoint returns 200

✗ token can be reused

? mobile UX unverified

DO NOT SHIP

AI fixes token reuse…

Human attests mobile UX…

rerunning evidence…

✓ auth regression tests

✓ reset endpoint returns 200

✓ token reuse rejected

✓ mobile UX attested

PROVEN ✓ · READY TO SHIP

1. AI says done 2. Proof finds gap 3. Fix + attest 4. PROVEN

Works with the tools you already use: Claude CodeCodexCursorGeminiPlaywrightJestpytestCypress

THE GAP

Writing code got faster. Proving “done” did not.

The weak point in vibe coding is no longer only implementation. It is knowing what was actually verified.

WITHOUT PROOF

Confidence is conversational

AI says “done”
Edge cases may be missing
Mobile / API can remain unchecked
You ask the same agent to review itself

CONFIDENCE: ASSUMED

✓

WITH TASHEV PROOF

Confidence is evidence

“Done” is an explicit contract
Checks execute independently
Unknown stays unverified
Ship decision is repeatable

CONFIDENCE: EVIDENCE

HOW IT WORKS

A verification layer between coding and shipping.

◎

Human task

Start from what the user actually asked for.

→

{ }

Acceptance contract

Define what must be true before the task counts as done.

→

⌁

Evidence

Run tests, files, HTTP checks and explicit human review.

→

✓

Verdict

PROVEN · PARTIAL · FAILED

HONEST OUTCOMES

Unknown is not success.

PROVEN

Every required criterion has evidence.

READY TO SHIP

PARTIAL

Nothing failed, but required evidence is missing.

VERIFY MORE

FAILED

At least one required criterion failed.

DO NOT SHIP

EVIDENCE, NOT VIBES

Use the tools your project already trusts.

Proof orchestrates evidence. It does not replace your test stack.

Command

tests · lint · build · Playwright · Cypress

File

exists · contains · regex · generated artifacts

HTTP

status · body · health · API · webhooks

Human

UX · visual · business · device · legal

contract.json task → evidence

{
  "task": "Add password reset",
  "criteria": [
    {
      "id": "auth-tests",
      "type": "command",
      "command": "npm test -- auth",
      "required": true
    },
    {
      "id": "health",
      "type": "http",
      "url": "http://localhost:3000/health",
      "status": 200,
      "required": true
    },
    {
      "id": "mobile-ux",
      "type": "manual",
      "required": true
    }
  ]
}

NOT ANOTHER TEST FRAMEWORK

Tests prove assertions. Proof proves the task.

Jest · pytest · Playwright

Did this test pass?

Excellent at executing technical assertions.

Tashev Proof

Is the human task proven complete?

Maps tests, HTTP, files and human review back to acceptance criteria.

GET STARTED

Install. Define “done”. Prove it.

Current release: v0.1.0 · MIT · zero runtime dependencies.

npm install -g https://github.com/tashev11/tashev-proof/releases/download/v0.1.0/tashev-proof-0.1.0.tgz

proof init --task "Add password reset" → proof run → proof ship

DOGFOODED

Proof had to prove Proof before shipping.

The v0.1.0 release includes its own generated Proof report.

View release evidence ↗

12PROVEN

0 failed 0 unverified ✓ ready to ship

TASHEV OPEN SOURCE

Three focused tools. Three different pains.

TashevOS

Orchestrate and remember AI development.

↔

Tashev Relay

Continue across agents, terminals and machines.

GitHub ↗

✓

Tashev Proof

Independently verify “done” before shipping.

GitHub ↗

STOP TRUSTING “DONE”

Make your next AI-built feature prove itself.

Open-source, local-first and designed to work with the stack you already have.

★ Star on GitHub Install v0.1.0