AegisPR V3 — Codegeist 2026

Inspiration 💡

Modern software development moves quickly, but pull request reviews often become a bottleneck. Developers must manually investigate code changes, identify regressions, evaluate security risks, and understand how changes affect the rest of a project.

AegisPR gives software teams a way to see which checks ran, which failed, and what evidence is still missing before they merge.

Don't just review a pull request. Verify the evidence.

What It Does 🚀

  • Bitbucket Cloud — Open the AegisPR panel on a pull request to see the exact revisions, changed files, static findings, assertion outcomes, and recent runs. PR creation and updates enqueue reviews. Reviewers can also run a review manually.
  • Base and head evidence — Forge reads the destination commit's JSON contract and evaluates the same assertions at both revisions. Supported checks report a scoped regression, passing assertions, or inconclusive evidence.
  • Jira Cloud — The PR identifies linked work items. An authorized reviewer can share a bounded evidence snapshot on the relevant issue and explicitly create a follow-up.
  • Rovo — Aegis Review Assistant reads the Jira issue and reviewer-shared report using three read-only actions. It explains actual evidence and identifies which assertions were unavailable.
  • V2 evidence workspace — The optional Next.js and FastAPI workspace retains the original dashboard, pinned GitHub diff review, signed receipts, prewritten fixture repairs and benchmark workflow. Reviewers can import a Bitbucket report. Optional model guidance is advisory.

How I Built It 🛠️

The two Forge apps use TypeScript and React. In Bitbucket, a PR event sends a job to Forge Async Events. The queue consumer fetches revision-pinned source and diffs from Bitbucket Cloud and stores a bounded report in installation-scoped Forge storage. The Jira companion reads work-item context and lets reviewers share a named snapshot. Rovo reads that snapshot under the requesting user's Jira permissions.

The Codegeist build began on November 16, 2026. It adds the Forge manifests, native Bitbucket and Jira panels, PR events and processing, declarative comparison, Jira context and sharing, Rovo actions, and the import that connects the evidence workspace to Forge. AegisPR's original GitHub workflow and V2 workspace are prior work. Reuse disclosure

Challenges I Faced ⚙️

PR event timing and revision drift required the worker to recheck the current state before saving a report. Forge product boundaries required separate Bitbucket and Jira apps. The reviewer explicitly shares a portable snapshot across that boundary, and each Jira reader sees it under Jira permissions. Evidence stays scoped to the checks that actually ran.

What I Learned 📚

The live demo compares real Bitbucket revisions. A 10% coupon check passed at base and failed at head while currency passed at both. A fresh PR automatically produced a passing report. A PR that moved the checked file produced an inconclusive report. Rovo named the failing check and explained the actual Jira requirement. The local benchmark ran 12 committed fixtures, reproduced eight seeded regressions, and abstained in three. Test report

What's Next 🌍

The preview verifies JSON contracts for same-repository Bitbucket changes. Further work includes more declarative check types and installation testing with an independent workspace administrator. The local workspace executes committed Python and JavaScript fixtures, with optional separately configured Docker and CUDA workers.

Final Vision

AegisPR helps developers make merge decisions from inspectable evidence. The Bitbucket panel, Jira context and Rovo explanation all refer to the same saved findings and verification outcomes.

Install and test the Forge apps · Source and release

Built With

Share this project:

Updates

Submission history