corvint-tests-accept

Experimental new-test assessment companion

Build with go build -o /tmp/corvint-tests-accept ./cmd/corvint-tests-accept. This separate local command implements the proposed new-e2e-test-acceptance-v0.md contract. It adds no Core command and executes only operator-approved disposable fixtures.

Prepare the closed request described by internal/testacceptance.Request: pinned clean product and test Git roots/commits/trees, hashes for tracked inputs, actual Node/CLI executable and entrypoint hashes, an owned loopback server command, declared new test identities, repeat count, environment label, and optional independently reviewed behavior-falsification plans with their separate executable and tool identity pins. TestNEAV0002ActualBrowserAssessment is a complete executable fixture. Never pass an author-selected plan or execute third-party code under a credentialed account.

Inspect before executing:

/tmp/corvint-tests-accept plan --plan /private/operator/request.json
/tmp/corvint-tests-accept accept --plan /private/operator/request.json \
  --approve-plan sha256:REVIEWED_DIGEST --new /absolute/test/root/new.spec.mjs \
  --repeat 2 --environment disposable-loopback > /private/operator/assessment.json

Repeat --new for multiple files. The flags must match the frozen request; changing inputs requires reviewing its new digest. plan reads and validates, without test execution. accept launches a sanitized self-worker for every repeat/probe/control. A test whose repeats disagree is also run alone twice; its assessment carries the order and isolation states. It emits JSON with a fixed Markdown body field (verdict, both revisions, environment, build, one row per test, reasons and unknowns) suitable for pasting into a draft PR; extract it with jq -r .body. No connector or outward write occurs. worker is an internal stdio protocol requiring the same explicit request digest and pin validation. Direct worker invocations with any environment key/value outside the fixed safe four-key environment are refused before execution; inherited credentials/runtime injection cannot enter this surface.

Current qualification: experimental rejection/blocked assessment only. Existing per-test Freshness and qualified baseline identity remain UNKNOWN. Stable asserting tests therefore stay blocked; actual surviving controls and flaky outcomes can be rejected. Missing controls block. V1-0556 tracks positive per-test freshness. Requested original/reversed file filters do not establish observed execution order. Cleanup observes known descendants at bounded intervals; fast detach can be missed. Imported SDK/browser closure, authenticated hook semantics and external outcomes are unqualified.

Focused tests:

GOTOOLCHAIN=local go test -count=1 -timeout 30m ./internal/testacceptance ./cmd/corvint-tests-accept
GOTOOLCHAIN=local go vet ./internal/testacceptance ./cmd/corvint-tests-accept
CORVINT_ACCEPTANCE_PLAYWRIGHT_CLI=/absolute/installed/playwright/cli.js \
  CORVINT_ACCEPTANCE_REPORT=/private/operator/live-assessment.json \
  GOTOOLCHAIN=local go test -count=1 -timeout 30m \
  -run '^TestNEAV0002ActualBrowserAssessment$' -v ./internal/testacceptance

Without the explicit installed Playwright CLI the live test skips with NOT_RUN; focused unit success does not imply browser qualification. The live fixture owns port 4173; a collision fails qualification and must not reap another process. It retains Node/browser versions, actual assertion kill/survival, alternating repeats, requested probes and explicit unknowns. It installs no dependency.