Performance regression gate

Benchmark key pages and APIs on every PR against the baseline, then block merges when latency, Lighthouse, or bundle thresholds regress.

performancebenchmarkinglighthousebundle-sizepr-gateregression

[ workflow / qa ]

Performance regression gate

Performance regressions often arrive through small changes. Cosmos replays production-representative load scenarios against each branch, compares Lighthouse, p95 API latency, and bundle size against a rolling baseline, and blocks regressions with a before/after breakdown and likely culprit.

07 nodes

05 edges

Trigger[trigger]

PR opened / updated

GitHub webhook

AI Agent step[deploy]

Build preview

Branch deploy

AI Agent step[bench]

Run benchmarks

Lighthouse · API latency · bundle size

AI Agent step[compare]

Compare to baseline

7-day rolling average

Decision

Regression detected?

Output / Result[pass]

Pass check

Green summary comment

Decision

Regression detected?

Output / Result[pass]

Pass check

Green summary comment

YES

Output / Result[block]

Block PR + report

Before/after table · culprit diff

Workflow prompt

Paste this into Augment to reproduce the workflow end-to-end.

Cosmos, on every pull request build a preview deployment and run the following benchmarks: Lighthouse performance score on the top 5 pages, p95 latency for the 10 highest-traffic API endpoints under 50 concurrent users, and total JS bundle size. Compare results against the last 7-day rolling baseline stored in the performance database. If any metric regresses by more than 10% (or bundle size grows by more than 5 KB), post a PR comment with a before/after table per metric, highlight the file diff most likely responsible, and set the PR check to 'fail'. Otherwise set to 'pass' and post a brief green summary.

← All Workflows