- Home
- Skills
- Testing & Debugging
- Branch Breaker — Pre-Merge QA & UX Review
Works with the AI Agents you already use
Branch Breaker — Pre-Merge QA & UX Review
Catch regressions before you merge. Stress-test branch changes, real user flows, and edge cases with your coding agent. Get reproducible bug reports, evidence, and a clear verdict on what was actually tested.
$9.99
Branch Breaker — Pre-Merge QA & UX Review
Example session with this skill installed
Use Branch Breaker to validate the demo-pagination-regression branch against main in this controlled Python API fixture. This demonstration deliberately contains a seeded defect; it is not a blind benchmark, customer case, or Agensi integration test. Run the existing tests, exercise the real loopback HTTP API, and verify whether collection-wide search finds eligible records outside the first 10-record page. Check pagination, combined text/status filters, input boundaries, and independent record reads. Compare the same probes against main. Preserve the feature code. Save a reproducible report, evidence, and local sharing drafts. Do not publish anything externally.
- Read your context and instructions
- Compiled the branch breaker
Branch Breaker — controlled demonstration
Verdict: FAIL — one confirmed P1 regression.
This run used a small synthetic Python API with an intentionally seeded defect. I knew the fixture contained a defect. This demonstrates execution and reporting; it is not a blind benchmark, a customer case, or an Agensi integration test.
Tested scope
Feature: demo-pagination-regression at 72b001c73891275ff75cc54f087b5d5842522799.
Base: main at e9c00aa9605423fbfc5185d98c634daa9b8e0ea7.
Runtime: Python 3.9.6 on macOS, using real loopback HTTP requests against separate feature and baseline server processes. No mocks or external services were used.
The collection contains 30 eligible synthetic records. The first page loads IDs 1–10. Record 26, Canary Delta, is active and initially outside that page. The documented contract requires text/status filtering across the full collection before pagination.
What worked
- All three existing happy-path tests passed on both revisions.
- Searching for the on-page
Canary Alpharecord returned ID 1. - A genuine no-match query and an excluded-status query returned empty results.
- An independent
GET /records/26returned the expected active record. - Invalid page, page size, and status inputs returned HTTP 400.
BB-001 — Search misses eligible records beyond the current page
Severity: P1. Attribution: introduced by the feature commit.
Reproduction: start the feature server with python3 server.py --port 0, then send GET /records?q=Canary%20Delta to its printed loopback port.
Expected: ID 26 on the first filtered page, with total: 1.
Actual feature response, HTTP 200:
{"items": [], "total": 0, "page": 1, "page_size": 10}
The same request on the base revision returned:
{"items": [{"id": 26, "name": "Canary Delta", "status": "active"}], "total": 1, "page": 1, "page_size": 10}
The missing result reproduced in three consecutive requests and remained missing after independently reading record 26. Searching on unfiltered page 3 exposed ID 26, confirming that results incorrectly depend on the pre-filter page window.
The same defect also corrupts collection totals, skips the second result of a paginated Canary search, and affects combined text/status filters. These are manifestations of one cause, not separate bugs.
Verified location: server.py:18–21 slices DATA before applying the filters and reports the sliced count. A client can therefore receive an empty result even though an eligible record exists.
Fix direction: filter the full eligible collection, compute the filtered total, then slice the matching records. No fix was applied.
Regression recommendation: add an API test with more than one page of data and an eligible match outside the initial page; assert that a first-page query returns that ID and the correct total. Add a multi-match next-page assertion to the normal local/PR checks.
Coverage and cleanup
The feature passed 8 of 20 HTTP checks and failed 12 checks associated with BB-001. The same 20 checks passed on the baseline. Check counts are coverage evidence, not a bug-detection score.
This was API-only coverage. There is no UI, native client, authentication, database, write endpoint, or LLM in this fixture. Screenshots of an application, visual/accessibility claims, and mobile coverage are not applicable. Query injection, XSS, CSS injection, JavaScript injection, and prompt injection have no interpreter/rendering/model sink here; none is claimed as security-tested.
Both server processes were stopped. The feature HEAD and all tracked-source fingerprints remained unchanged. Local evidence, a complete technical report, and Slack/Jira sharing drafts were saved. Nothing was posted externally.
Connects securely to your tools. The creator never sees your data.
About this skill
Branch Breaker turns your coding agent into a skeptical QA reviewer for your next merge. Use it after a feature, fix, or refactor to investigate what could break — and give your team findings they can reproduce.
Test beyond the happy path
The workflow traces changed code to affected consumers, runs the project’s available checks, and exercises relevant feature journeys. It probes input boundaries, permissions, persistence, failure recovery, accessibility, and downstream regressions. Search and filter changes get an explicit check for records beyond the currently loaded page.
Get a report you can act on
- Reproducible findings: severity, exact steps, expected versus actual behavior, and supporting evidence.
- Honest coverage: the tested revision or build, affected areas, checks performed, and anything that could not be verified.
- A clear verdict: FAIL, INCOMPLETE, or PASS (scoped), with next steps and focused regression-test recommendations.
- Review-ready artifacts: local reports, real screenshots when available, and drafts you can share with your team.
Built for branch reviews
Use it before merging a feature, investigating a regression, or reviewing a changed user journey. It separates confirmed defects from UX suggestions and product questions, so you can prioritize with evidence.
The workflow covers web apps, mobile web, native Android/iOS, APIs, CLIs, workers, and libraries when the necessary runtime and tools are available. It discovers and uses your project’s existing tools.
Start with a concrete request
Load the skill in a compatible coding agent with access to your checkout and test environment. Then ask:
Use Branch Breaker to review this branch against main before merge. Exercise the changed flows, check edge cases and downstream effects, and return reproducible findings with coverage gaps.
Requirements and scope
You need Git, a coding agent with shell and filesystem access, project dependencies, and an authorized test environment. Real web UI checks need browser automation; native checks need a compatible build and device or simulator. Missing tools are reported as coverage gaps.
Branch Breaker writes reports under .tmp/branch-breaker/ and preserves product code. External posting requires your explicit authorization.
Packaged skill release: 1.7.0 · MIT licensed by GustavoEmmel. The package includes validation notes. Agensi delivery has not yet been validated end to end.
Coverage depends on the coding agent, project setup, and tools available. Native Android/iOS checks require a compatible build and device or simulator; browser viewport emulation covers mobile web only. Missing access is reported as incomplete coverage. The public demo uses a controlled Python API fixture with an intentionally seeded defect and does not validate every target stack. Delivery through Agensi has not yet been validated end to end.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 5 days ago
Requires
Permissions
Reads the target checkout; runs project build/test commands and controlled probes in an authorized test environment; writes local QA reports, sanitized evidence, and sharing drafts. Uses browser/network access where the target requires it. Product code is preserved. External posting requires explicit authorization.