The trust layer for autonomous testing
VIGILIS
Self-healing Playwright tests you can verify — every fix sealed in a signed, tamper-evident receipt, and it never masks a real bug.
One loop, four behaviours
Generate
Point it at a URL → it explores the app and writes a runnable Playwright spec.
Gate
The spec runs in CI as a required check — failing tests block the deploy.
Triage
On failure it classifies the cause: real bug vs DOM drift vs flake.
Heal
For drift it rewrites the locator, verifies green, and opens a PR — never masking real bugs.
One core, two consumers
Define the QA tools once in @argus/core; expose them as a CLI and an MCP server.
Why Vigilis
AI writes real tests
It explores your app and emits runnable Playwright specs — not brittle boilerplate.
Failing tests block deploy
Vigilis runs as a required GitHub check, so regressions never reach production.
Self-healing PRs
DOM drift is fixed automatically and verified green before a PR — real bugs are never hidden.
Signed provenance receipts
Every heal is wrapped in a Treeship session: an auditable, signed chain of artifacts.
Drive it from Claude
The same tools ship as an MCP server — generate, triage and heal straight from your agent.
One core, two consumers
Define the QA tools once; reuse them across the CLI and the MCP server with zero drift.
Self-healing QA you can audit.
Every run is sealed inside a signed Treeship session — each tool call, decision and verdict captured in a tamper-evident chain, signed by an independent notary. You get a receipt, not a black box.
Never a false green. Vigilis heals cosmetic drift but refuses to heal a real bug — and the receipt proves which call it made, and why.
Real, live receipts — public, no login, independently verifiable.
$ argus heal --spec checkout.spec.ts
triage → dom-drift (locator stale)
heal → rewrote getByRole('button')
verify → 12 passed (green)
pr → #42 opened
✓ verified · chain intact · independently signed