Truly Done v0.1.0

Truly Done 0.1.0

Checks that an AI agent's "done" is truly done: real clicks in headless Chrome on your running app, then PASS, FAIL or CANNOT TELL with evidence. No telemetry, no cloud, no account, Python standard library only.

• verify.py run: drives a running web app in headless Chrome from a JSON flow (goto, reload, click, type, press,
  select, check, hover, wait, wait_for, expect) and prints VERDICT PASS, FAIL or CANNOT TELL.
• 13 claim kinds: text, no_text, count, visible, hidden, value, url, title, dialog, request, no_request,
  no_console_errors, no_failed_requests.
• A loopback base URL may be written without a scheme: --url localhost:3000 means http://localhost:3000.
• verify.py snapshot (controls on a page), check (validate a flow), gate (exit 0 only when every report passed).
• Report per flow: report.json, report.md, final.png. Secrets typed from {"env": "NAME"} are redacted.
• Guard rails in code: loopback by default (--mine for any other host), writes to other origins blocked,
  --offline, a fresh browser profile per run, no telemetry, no update check, no account.
• GitHub Action (uses: maumoya/truly-done@v0.1.0) with a self-test: one passing run, one failing run.
• Benchmark bench/run.py: 9 planted bugs and 3 healthy variants of a fixture app.

Use it in CI

- uses: maumoya/truly-done@v0.1.0
  with:
    flows: flows/*.json
    url: 127.0.0.1:3000

Verify

git verify-tag v0.1.0
slsa-verifier verify-artifact truly-done-v0.1.0.zip --provenance-path truly-done-v0.1.0.intoto.jsonl --source-uri github.com/maumoya/truly-done --source-tag v0.1.0