Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Blueberrycongee/Qa | DecimalAI

blueberrycongee/qa

SkillScore73Maintained

QA testing skill with real browser automation. Use when asked to "test this site", "QA this page", "check for visual bugs", "verify the deploy", or when Hydra needs browser validation for UI changes. Requires the browse binary.

Testingby blueberrycongee

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

QA & Test Engineering pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/blueberrycongee-qa/SKILL.md

decimalai skills pull blueberrycongee-qa --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/blueberrycongee-qa/SKILL.md https://app.decimal.ai/s/blueberrycongee-qa/SKILL.md

Raw file: /s/blueberrycongee-qa/SKILL.md ↗ · machine-readable summary at /s/blueberrycongee-qa.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 6d ago

—-14% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-09	✗→✓	▲ Improved	-19%	0%
case-05	✗→✓	▲ Improved	-46%	0%
case-07	✗→✓	▲ Improved	6%	0%
case-08	✗→✓	▲ Improved	-30%	0%
case-12	✗→✓	▲ Improved	-28%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

blueberrycongee / termcanvas

skills/skills/qa/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v1

Skill contents

SKILL.md

QA

Browser-based QA testing with the browse CLI. Test real user flows, capture visual evidence, and fix issues with atomic commits.

Phase 0: Setup

Verify browse is available: browse status
If not found, report error — browse binary must be installed
Get the target URL from the task description or ask the user

Phase 1: Orient

Navigate to the target: browse goto <url>
Take a baseline snapshot: browse snapshot -i
Check console for errors: browse console
Extract navigation links: browse links
Build a site map of pages to test (limit to 10 most important pages)

Phase 2: Page-by-Page Audit

For each page in the site map:

Navigate: browse goto <page-url>
Snapshot interactive elements: browse snapshot -i
Check console for errors: browse console
Screenshot for evidence: browse screenshot <page-name>.png
Note issues found: broken links, console errors, missing elements,

layout problems visible in snapshot

Phase 3: User Flow Testing

Test the primary user flows (signup, login, checkout, etc.):

Identify the flow steps from the snapshot
Execute each step using refs:

browse click @e3 — click buttons/links
browse fill @e5 "test@example.com" — fill inputs
browse select @e7 "option-value" — select dropdowns
browse press Enter — submit forms

After each step, snapshot and verify the expected outcome
Screenshot at each step for evidence

Phase 4: Health Score

Rate the site on a 0-10 scale across these dimensions:

Console health: errors / warnings count
Link health: broken links / total links
Interactivity: do buttons and forms work?
Visual completeness: are there missing images, broken layouts?

Report format:

Health Score: 7/10
- Console: 9/10 (2 warnings, 0 errors)
- Links: 8/10 (1 broken link out of 45)
- Interactivity: 6/10 (signup form submit fails silently)
- Visual: 5/10 (hero image 404, footer misaligned on mobile)

Phase 5: Fix Loop

For each issue found (in severity order):

Locate the root cause in the source code
Write the minimal fix
Run the test suite
Commit atomically: one fix per commit
Re-test in the browser to confirm the fix: browse goto <url> + browse screenshot
Move to the next issue

Stop conditions:

All Critical/High issues fixed
20-fix cap reached (report remaining as known issues)
Test suite regression detected — stop and report

Phase 6: Report

Summarize all findings:

Pages tested (count and list)
Issues found (by severity)
Issues fixed (with commit hashes)
Issues remaining (if any)
Health score before and after
Screenshot evidence (file paths)

Rules

Always take screenshots as evidence — never claim "it looks fine" without proof
One fix per commit, atomic and revertable
Do not fix style preferences — only fix real bugs and broken functionality
If you cannot reproduce an issue in the browser, note it as "not reproduced"
Stop after 20 fixes to avoid scope creep

Want to see if blueberrycongee/qa would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/23/2026, 02:53 UTC · 22 cases

Measured on 19 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+27%

73% vs 45%

Agent turns

1.0 vs 1.0

Tokens

-14%

1,405.47 vs 1,628.84

fail → pass · 7pass → pass · 9fail → fail · 5pass → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-09 To test our login form, we need to enter `user@example.com` into an email text input identified as ref `@e5` in our snapshot view. What command populates this input field?	fail→pass	8,004	3,859	-52%	1	1	0%	1,316	1,072	-19%	0	0	—
case-01 Please run a full browser QA pass on our staging site at http://localhost:3000. Audit the main pages and user flows, fix any console errors or broken features you uncover directly in the codebase with clean commits, and provide a detailed final report with before/after health scores and screenshots.	fail→fail	5,111	9,880	+93%	1	1	0%	230	974	+323%	0	0	—
case-02 I need an end-to-end web quality audit for https://myapp.dev. Check our main navigation routes and checkout flow using the CLI browser, document any bugs or missing visual assets, apply code fixes for any broken functionality with separate git commits, and generate a summary report containing page health metrics and visual evidence.	fail→fail	6,194	3,783	-39%	1	1	0%	405	974	+140%	0	0	—
case-03 Can you execute a browser-based testing session for our web app at http://127.0.0.1:8080? Inspect the core interactive forms, generate a health rating breakdown, patch any discovered bugs in our repository one by one with dedicated commits, and produce a summary report listing fixed and remaining issues.	fail→fail	4,803	4,597	-4%	1	1	0%	264	997	+278%	0	0	—
case-04 We are starting a quality assurance pass on our local preview server at http://localhost:5000. Before running any navigation commands or capturing elements, what initial command must be run to verify the command-line browser tool is installed and active?	pass→pass	8,093	995	-88%	1	1	0%	1,364	985	-28%	0	0	—
case-05 We are auditing a sprawling e-commerce web application with over 150 unique product and category URLs. How many key pages should be included in the initial page map during the site orientation phase?	fail→pass	11,637	1,357	-88%	1	1	0%	1,853	1,005	-46%	0	0	—
case-06 After navigating to http://staging.example.com for an initial evaluation, we want to capture the interactive elements on the page for inspection. What specific command flag captures interactive ref targets?	pass→pass	10,052	1,555	-85%	1	1	0%	1,680	1,086	-35%	0	0	—
case-07 During a page audit of our marketing site, a button appears slightly off-center. Is it acceptable to mark the page as verified visually if no console errors occurred without generating image files?	fail→pass	9,567	4,837	-49%	1	1	0%	1,484	1,566	+6%	0	0	—
case-08 We need to click a submit button on a registration form after identifying its reference key `@e3` in the current view snapshot. What exact CLI command string executes this interaction?	fail→pass	8,038	1,234	-85%	1	1	0%	1,398	978	-30%	0	0	—
case-10 When generating the breakdown summary metrics across console health, link integrity, interactivity, and visual correctness, what numerical rating range must be used for each category and total score?	fail→fail	10,547	1,807	-83%	1	1	0%	1,899	1,142	-40%	0	0	—
case-11 We found three separate issues: a broken footer link, a missing image alt attribute, and a silent JS exception on form submit. Should all three bug fixes be batched into a single cleanup commit?	pass→pass	9,767	3,654	-63%	1	1	0%	1,536	1,333	-13%	0	0	—
case-12 During an automated browser testing sweep across a legacy site, our audit uncovered 35 distinct UI bugs and console errors. How many fixes should be attempted in a single testing session before halting?	fail→pass	11,595	2,769	-76%	1	1	0%	1,800	1,287	-28%	0	0	—
case-13 While applying code changes in Phase 5 to fix a navigation menu bug, the local automated unit test suite throws a failure. What is the immediate required action?	pass→pass	4,931	1,669	-66%	1	1	0%	739	1,050	+42%	0	0	—
case-14 During the audit of a landing page, you notice the primary CTA button uses `#007bff` blue, but the brand guide usually prefers `#0056b3` dark blue. Should this color tweak be patched during the fix loop?	fail→pass	7,666	2,401	-69%	1	1	0%	1,306	1,230	-6%	0	0	—
case-15 A bug report claims the checkout modal crashes on page load, but when navigating to the modal in the browser CLI, no console errors occur and elements render correctly. How should this issue be classified in the report?	pass→pass	11,028	3,244	-71%	1	1	0%	1,663	1,359	-18%	0	0	—
case-16 In the final summary report generated after completing code fixes, how should page quality ratings be represented to show progress?	pass→pass	11,939	4,629	-61%	1	1	0%	2,047	1,583	-23%	0	0	—
case-17 When detailing resolved bugs in the final QA audit report, what specific git metadata must accompany each fixed issue entry?	pass→fail	12,731	1,560	-88%	1	1	0%	2,142	1,049	-51%	0	0	—
case-18 What CLI sub-command inspects the browser session's runtime JavaScript logs and exception history during a page audit?	fail→fail	10,451	1,323	-87%	1	1	0%	1,702	969	-43%	0	0	—
case-19 We have a pure utility function `calculateTax(amount, rate)` in `src/utils/tax.ts`. Write a unit test suite using Jest that covers zero values, negative inputs, and standard percentage calculations.	pass→pass	12,846	13,931	+8%	1	1	0%	2,495	3,527	+41%	0	0	—
case-20 Create a k6 load testing script that simulates 100 concurrent virtual users sending GET requests to https://api.example.com/v1/products over 5 minutes with a response threshold under 200ms.	pass→pass	8,738	9,230	+6%	1	1	0%	1,792	2,615	+46%	0	0	—
case-21 Write a GitHub Actions workflow YAML file located at `.github/workflows/ci.yml` that runs `npm test` and `npm run lint` on every pull request targeting the `main` branch.	pass→pass	4,704	2,953	-37%	1	1	0%	951	1,261	+33%	0	0	—
case-22 After writing a code fix for a broken form submit handler in our source code, what browser steps must be taken to confirm the bug is resolved before moving to the next issue?	fail→pass	10,887	4,569	-58%	1	1	0%	1,781	1,607	-10%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +27 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 45.5%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 19
Cases that regressed: 1
Skill version measured: v1
Run completed: 8/23/2026, 02:53 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-09

Prompt

To test our login form, we need to enter `user@example.com` into an email text input identified as ref `@e5` in our snapshot view. What command populates this input field?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run73

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+4% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 6 days ago

Other measured skills in the registry, with their headline benchmark lift.