Skip to content
Wildan.
Back to work

Hackathon Prototype · Offline

SpecHeal

A recovery cockpit that separates safe test healing from real product bugs.

Role
Team Lead · Full-Stack & Product Developer
Year
2026
State
offline

A test turning green does not always mean the product is correct. I led SpecHeal around that constraint, combining Playwright failure context, specification guardrails, structured AI verdicts, deterministic browser checks, controlled patches, and verified reruns.

The problem

Many self-healing tools optimize for turning a red test green, which can create a false green when the product is actually broken or intended behavior has changed.

Built for

QA engineers reviewing UI test failures

Product teams that need clear recovery decisions

Engineers exploring guarded AI-assisted developer tooling

How it works

  1. 01

    Capture Playwright failure context and apply an OpenSpec guardrail.

  2. 02

    Produce a structured AI verdict instead of an unchecked locator replacement.

  3. 03

    Validate candidate changes in the browser before applying a controlled patch.

  4. 04

    Require a verification rerun and persist an audit/Jira-ready handoff.

What I built

Led the team and shaped the trustworthy-recovery product thesis.

Built the full-stack recovery cockpit and product workflow.

Integrated failure capture, structured verdicts, browser validation, patches, and reruns.

Connected technical outcomes to a persistent audit trail and handoff.

SpecHeal was a collaborative hackathon MVP. The team placed second at Refactory Hackathon 2026; the result is attributed to the team, while Wildan's role remains scoped to leadership, product definition, and full-stack delivery.

Key decisions

Optimize against false greens

Not every selector failure is safe to heal.

Tradeoff: Guarded recovery is slower than blind replacement but protects test meaning.

Require a verification rerun

A replacement locator does not prove the intended user state works.

Tradeoff: Each accepted repair needs additional browser execution and result storage.

Results & sources

MY ROLEAS OF 2026-07-15

Wildan served as Team Lead, Full-Stack Developer, and Product Developer.

Canonical repository
PROJECT RECORDAS OF 2026-07-15

The repository implements an evidence-to-verdict-to-rerun loop for seeded Playwright scenarios.

Canonical repository
PROJECT RECORDAS OF 2026-07-15

The historical hackathon demo is currently unavailable.

Canonical repository