Inspiration

A Jira ticket can contain beautiful acceptance criteria and still have no evidence that anything was tested. SpecGuard addresses that dangerous gap between written requirements and verified work.

What it does

SpecGuard is a read-only Jira Cloud issue panel that extracts acceptance criteria, checks whether key specification sections exist, and highlights missing test and review references. It never turns a written acceptance criterion into a fictitious PASS, never alters a Jira issue, and never makes a release decision.

On our live, fictional example SGD-1, SpecGuard identifies three criteria, all NOT VERIFIED, and reports specification completeness 3/4. On SGD-2, it finds no structured acceptance criteria and reports 0/4 completeness with actionable specification gaps. The Refresh analysis control rereads the current Jira issue through the signed-in user's permissions.

How we built it

The app is built on Atlassian Forge using a jira:issuePanel, @forge/react, @forge/bridge and @forge/resolver. Its backend reads Jira issue descriptions with asUser().requestJira, processes Atlassian Document Format, and runs deterministic JavaScript checks. We deliberately avoid external AI services, cloud API credentials, and speculative test verdicts.

The real application was deployed and installed in the Forge production environment on Jira Cloud. Local regression verification: 35 passing tests.

Challenges

Distinguishing an explicitly written criterion from actual test evidence was the critical design challenge. We also handled incomplete Jira specifications, nested document structure, permissions, conservative status labels, and deployment to a real production Forge environment.

Accomplishments

  • Working, deployed Forge issue panel, not a mock-up.
  • Evidence-first NOT VERIFIED status instead of fabricated green checkmarks.
  • Explainable completeness and missing-evidence findings.
  • A refresh action that does not write to Jira.
  • Authentic 74-second Jira Cloud demonstration using two fictional tickets.

What we learned

A useful QA assistant need not pretend to know what it cannot verify. Showing the limits of available evidence makes human review faster and more trustworthy.

What's next

Improve evidence-link discovery, add configurable specification templates, and validate the workflow with more representative issue documents. Any future evidence verification should remain explicit, permission-aware, and auditable.

Links

Scope: SpecGuard flags evidence gaps. It does not certify that tests were executed or that software is ready for release.

Share this project:

Updates

Submission history