performance-regression

Runtime evidence that helps agents trace, profile, and burn down hotspots in application and native code, GPU kernels, and inference stacks.

Build trustworthy pytest-benchmark matrices, collect repeated runs, and detect performance regressions

Claude Code plugin for QA/perf engineers: paste a curl, HAR, Postman or OpenAPI request → k6 load test → runs locally or in your CI → markdown/wiki report → regression verdict vs. the previous run, computed by a deterministic script, not the LLM. No Grafana Cloud needed.

Dart/Flutter CLI for static rebuild-scope analysis, structural mutation checks and profile-mode rebuild measurement. Since 0.8.0, spm screen flags the likely direction of a UI change's rebuild cost, statically, without running the app.

Compare OpenTelemetry traces across releases, gate latency and error regressions in CI, and inspect the evidence through MCP. Built with Kotlin and SQLite.

LLM agent that turns two Android Perfetto traces and a git range into a cited diagnosis: the regressed metric, the culprit commit, and the trace rows that prove it. Every claim is re-verified by code; measured on planted regressions.

Open-source inference cost doctor for self-hosted LLM serving: $/1M tokens from Prometheus /metrics, API price comparison, and a CI gate (PASS/FAIL/INCONCLUSIVE) for performance and cost regressions. vLLM, SGLang, TensorRT-LLM, Dynamo.

Continuous benchmarking and performance-regression tracking. Runs benchmarks, records baselines in a git branch, renders trend dashboards, and posts non-blocking baseline-vs-PR comparisons.

Production-grade guides and runbooks for query plan baseline tracking, performance regression detection, and CI/CD gate automation — for database SREs and platform teams.

Compare OpenTelemetry traces from before and after a change: finds N+1 queries, retry storms, fan-out growth and new calls on the critical path that tests and latency dashboards miss.

Example of automatically generated performance analysis reports — comparing builds, identifying regressions, and producing decision-ready conclusions from performance test data.

Evidence-backed performance regression guardrails for pull requests, with protected verification and bounded Codex repair workflows.

Claude Code skill to benchmark code, save baselines, compare stacks, and catch slowdowns. Built on affaan-m/ECC by @affaan-m.

Profile a pytest suite, script, or callable and comment a CPU + memory hotspot diff on every pull request.

JMeter-based performance testing solution built under a critical production performance incident, demonstrating stateful workload modelling, observability, capacity testing, and iterative performance engineering.

Continuous benchmarks for the Brightway LCA stack, with a correctness check beside every timing

Compare Rust benchmark results across pull-request revisions

Agent swarm that bisects vLLM on a real A100, names the commit that slowed your Llama endpoint, and proves it with statistics. Open weights end to end (Llama-3.2-1B + muse-spark-open).

Planted performance regressions in Google's JetNews sample (Jetpack Compose) with Perfetto captures: startup I/O, allocation storm, main-thread sleep, layout thrash, listener leak. Ground truth for evaluating Android performance tools.