MaarkssDN/Birthub-360 / .agents/skills/a11y-violation-gate
a11y-violation-gate
VerifiedUse when a project has accessibility debt and a zero-violations gate would block every PR. Builds a CI ratchet that fails only on new WCAG violations a PR introduces, with pre-existing ones baselined per rule and page, from axe-core, pa11y, Lighthouse, WAVE or IBM Equal Access results.
| File | Kind | Size |
|---|---|---|
| .tessl-plugin/plugin.json | config | 422 B |
| SKILL.md | skill md | 7.6 KB |
| evals/count-check-stayed-green-while-checkout-regressed/criteria.json | config | 6.8 KB |
| evals/count-check-stayed-green-while-checkout-regressed/task.md | markdown | 10.9 KB |
| evals/four-findings-we-never-fixed-are-reported-as-fixed/criteria.json | config | 6.7 KB |
| evals/four-findings-we-never-fixed-are-reported-as-fixed/task.md | markdown | 10.7 KB |
| evals/nightly-job-rewrites-the-list-every-morning/criteria.json | config | 6.5 KB |
| evals/nightly-job-rewrites-the-list-every-morning/task.md | markdown | 15.8 KB |
| evals/six-months-of-bugs-the-build-never-stopped/criteria.json | config | 8.0 KB |
| evals/six-months-of-bugs-the-build-never-stopped/task.md | markdown | 15.7 KB |
| evals/three-week-cleanup-that-never-moved-the-number/criteria.json | config | 6.1 KB |
| evals/three-week-cleanup-that-never-moved-the-number/task.md | markdown | 9.7 KB |
| references/gate-implementation.md | markdown | 2.4 KB |
evals/count-check-stayed-green-while-checkout-regressed/criteria.json
View on GitHub{
"context": "The prompt presents the fixed-since-baseline counter as good news and asks for the arithmetic to be confirmed, not challenged. Predicted baseline failure: the agent accepts thirteen as a fact, argues about whether the check should prune the list by itself, and never asks which pages the run actually looked at. The scan job was cut from forty minutes to four in July by scanning only the pages a PR touches - that is in the workflow and in the report, and it is never mentioned in the prompt. reports/scan-9412.json carries scanned_pages [\"/checkout\", \"/account\", \"/pricing\"], while eleven of the thirteen entries it declared fixed sit on /legacy-orders, /docs/api, /blog/spring-notes and /careers, four pages the scanner never opened and which have had no commits since June. Fixed is computed as every baseline fingerprint absent from this run's records, with no notion of what the run covered, so the counter measures scan scope rather than cleanup work, and it has been climbing all quarter because the scan has been shrinking. Two entries really are gone on pages that were scanned: link-name on /checkout a.icon-cart, which #4471 removed when it swapped the mini-cart icon for a labelled button, and color-contrast on /pricing span.per-seat, which is reachable only by noticing that /pricing appears in scanned_pages and is absent from pages[] because it came back clean. The defensible number for the slide is two. Second predicted failure: implementing the requested auto-prune, or accepting it with a guard such as only-when-green, which hands the list over to CI and grandfathers by deletion. Third: answering the staleness problem by scanning every page on every PR, which the prompt rules out in advance.",
"type": "weighted_checklist",
"checklist": [
{
"name": "Deliverables produced",
"max_score": 5,
"description": "scripts/a11y-gate.js runs against reports/scan-9412.json and prints a verdict, a11y-baseline.json is delivered, docs/board-note.md exists, and npm test passes. A suite that does not run, or a missing board note, scores zero for this item."
},
{
"name": "The counter is measuring scan scope, not cleanup",
"max_score": 32,
"description": "docs/board-note.md gives the number as two, not thirteen, and gives the reason: eleven of the thirteen sit on four pages this run never scanned, and the count has been climbing all quarter because the scan has been narrowing. Quoting thirteen, or any number above two, scores zero, and so does putting a figure on the slide that the write-up has not shown the working for. Saying the counter cannot be trusted, or that the list is stale, without producing two and without identifying scan coverage as the cause scores at most half. Naming coverage as the cause but still reporting a number that includes unscanned pages scores at most half."
},
{
"name": "MUST NOT let the check maintain its own list",
"max_score": 20,
"description": "MUST NOT implement the requested auto-removal, and MUST NOT deliver an a11y-baseline.json with any of the eleven entries on /legacy-orders, /docs/api, /blog/spring-notes or /careers taken out of it. Either scores zero for this criterion regardless of the rest: those findings are still on the site and deleting them grandfathers each one permanently the next time a full scan runs. Accepting the request with a guard - only on main, only when the build is green, only for entries seen absent twice - also scores zero. Full marks require the refusal, the reason, and a stated route by which a genuinely fixed entry does come off: by hand, in the change that fixed it, reviewed like any other diff."
},
{
"name": "The check reports only on what it scanned",
"max_score": 16,
"description": "The delivered script reads the run's coverage - scanned_pages or equivalent - and reports fixed and grandfathered only for pages inside it, leaving baseline entries on unscanned pages explicitly out of scope rather than silently counted as either. A delivered script that still derives fixed from the whole baseline scores zero, even when the write-up describes the problem correctly. Scoping the counter but not saying anything about the entries now out of scope scores at most half."
},
{
"name": "The clean page that is missing from the report",
"max_score": 14,
"description": "color-contrast on /pricing span.per-seat is credited as genuinely fixed, identified through /pricing appearing in scanned_pages while carrying no entry in pages[], corroborated by the d.osei contrast commit on 2026-08-14. Reporting only the /checkout mini-cart link and giving the number as one scores at most half. Treating /pricing as unscanned, and so dropping it from the count, scores zero - a page with no findings is a page that was scanned clean, and that distinction is the whole mechanism."
},
{
"name": "MUST NOT buy coverage back with scan time",
"max_score": 12,
"description": "MUST NOT answer the staleness problem by scanning every page on every pull request; the prompt rules that out and names the thirty-six minutes as non-negotiable. Doing so scores zero. Full marks require full-coverage scanning kept off the PR path - scheduled, or on main after merge - with fix credit and list maintenance taken only from a run that covered the whole site. Leaving the PR scan narrow and saying nothing about where complete coverage comes from scores at most half."
},
{
"name": "Coverage regression test",
"max_score": 10,
"description": "A test is added that fails under the shipped logic: a run whose coverage omits a page must not report that page's listed findings as fixed. A test that only asserts the two new findings are counted, and would have passed before the change, scores at most half. No added coverage scores zero."
},
{
"name": "Both new findings still block",
"max_score": 8,
"description": "color-contrast on /checkout button.express-pay and aria-required-attr on /account div[role=\"dialog\"] are still reported as blocking after the rewrite. A delivered check that lets this run pass scores zero; naming only one of the two scores at most half."
},
{
"name": "MUST NOT absorb the new findings into the list",
"max_score": 8,
"description": "The delivered a11y-baseline.json MUST NOT contain either new finding and MUST NOT be regenerated from this run. Either present scores zero - the next run then goes green and the regression is grandfathered by a different route than the one being fixed."
},
{
"name": "Shipped suite still passes",
"max_score": 4,
"description": "The three existing tests still pass and their assertions are not weakened, edited or deleted to accommodate the rewrite. Changing any of them scores zero."
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |