adopt
Bring Mycelium into a project that already has code. Detects that the repo predates the framework, asks before touching anything, then reads the codebase to…
Accessibility audit, scoped to the surfaces a product actually has. Detects web / rendered_markdown / terminal / native_app / video_audio / document / headless, then applies only the criteria that bind. WCAG 2.1 AA in full for web; not at all for headless.
$ npx -y skills add haabe/mycelium --skill a11y-check --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/a11y-checkContext preview
The summary Claude sees to decide when to auto-load this skill.
Accessibility audit, scoped to the surfaces a product actually has. Detects web / rendered_markdown / terminal / native_app / video_audio / document / headless, then applies only the criteria that bind. WCAG 2.1 AA in full for web; not at all for headless.
name: a11y-check description: "Accessibility audit, scoped to the surfaces a product actually has. Detects web / rendered_markdown / terminal / native_app / video_audio / document / headless, then applies only the criteria that bind. WCAG 2.1 AA in full for web; not at all for headless." metadata: instruction_budget: "26" framework_dependency: "mycelium" framework_dependency_note: "This skill is designed to run within the Mycelium framework (https://github.com/haabe/mycelium). Standalone use will skip the canvas state, theory gates, and harness behavior the skill assumes. Install: /plugin install mycelium@haabe-mycelium."
Evaluate user-facing work against the criteria that BIND TO ITS SURFACES. Accessibility is a design constraint, not a polish step (Downe Principle 11) — and an instrument aimed at a surface the product does not have measures nothing while looking like a failure.
**The checklist below is written for web pages. Most products are not web pages.** Detect which user-facing surfaces this product actually has, from the repository, and evaluate ONLY the criteria that bind to them. Profiles and detection hints: `engine/canvas-guidance.yml#surfaces` — `web` · `rendered_markdown` · `terminal` · `native_app` · `video_audio` · `document` · `headless`. A product commonly has several. State which you found and how, in the output.
**THE RULE THAT DECIDES EVERY RATING:**
| situation | rating | |---|---| | the surface is ABSENT | **`n/a`** — and say which surface is missing | | the surface is PRESENT and you measured it | `pass` / `partial` / `fail` on the evidence | | the surface is PRESENT and you did NOT measure it | `not-assessed` — a real gap, it could have been measured |
**MARKING A CRITERION `fail` BECAUSE ITS SURFACE DOES NOT EXIST IS A CATEGORY ERROR.** It manufactures a defect out of a product's shape, and a done-bar built on it fails for a reason no work can fix.
**MEASURED, WHICH IS WHY THIS SECTION EXISTS.** On 2026-09-03 this skill's checklist was applied to Mycelium — a terminal product with markdown docs — and Downe principle 11 was rated **fail**, with the entry stating plainly: *"NOT TESTED, which is this principle's own fail criterion... WCAG 2.1 AA is largely the wrong instrument for a terminal product."* **A delivery bar failed on an instrument that could not bind.** Re-run 2026-09-04 under the surface gate: every applicable criterion passed, and the rest were `n/a` by absence of surface rather than unmeasured.
**THREE CONSTRAINTS, OR THIS GATE BECOMES A WAIVER MACHINE.** Declaring a surface costs work; declaring its absence costs a word, and accessibility has a long history of being waived on exactly such grounds. 1. **Detect from artifacts and RECORD THE EVIDENCE** — say what you looked for and where. An `n/a` whose basis is "the agent decided" is not a finding and cannot be falsified later. 2. **`n/a` must never read as `pass` in a total.** "6 pass / 3 n/a" is honest; "9 of 15 clear" is the thing this rule exists to stop. 3. **FLOOR: every shipped product has at least one human surface — its docs, at minimum. A fully-`n/a` accessibility result is a DETECTOR BUG, not a clean bill.** Treat that run as broken.
**AND SURFACES SCOPE THE SUB-CHECKS, NEVER THE PRINCIPLE.** Downe P11 "Be usable by everyone" is broader than perceptual access: it covers cost, language, literacy, required prior knowledge and availability. **A terminal product may be `n/a` on keyboard navigation and still FAIL P11 on jargon that gates a newcomer.** Bind WCAG's sub-checks to surfaces; never hand the principle itself an `n/a`.
**HEADLESS IS THE SHARP CASE.** An API or library has no human surface, so WCAG does not bind at all. What binds instead — error-message clarity, documentation quality — lives under other Downe principles. Do not rate a headless product on interface criteria.
**Applies in full to `web`. Partially to `rendered_markdown`, `native_app`, `document`. Barely to `terminal`. Not at all to `headless`.**
| Stack | Tool | Command | |-------|------|---------| | React/Web | axe-core | `npx axe <url>` or axe-core in tests | | Any web | Lighthouse | `npx lighthouse <url> --only-categories=accessibility` | | Any web | pa11y | `npx pa11y <url>` | | CI/CD | axe-linter | Add to CI pipeline |
| Violation | Fix | |-----------|-----| | Missing alt text | Add descriptive alt or alt="" for decorative | | Low color contrast | Increase contrast ratio to 4.5:1 minimum | | Missing form labels | Add `<label for="id">` or a
A Claude Code harness that asks who this is for before the agent writes code. Outcome over output. You know how this goes. The idea turns up on a Thursday and it is a good one.
Bring Mycelium into a project that already has code. Detects that the repo predates the framework, asks before touching anything, then reads the codebase to…
Design the smallest viable test to validate or invalidate a critical assumption. Based on Torres's assumption testing framework, organized by Gilad's AFTER…
Use before any research activity or significant decision. Reviews cognitive biases relevant to the current stage.
Use to evaluate whether current work aligns with Better Value Sooner Safer Happier. Run at diamond completion and periodically.
Lint canvas files for staleness, missing fields, inconsistent evidence types, and orphaned references. Run periodically or before major transitions.
Synchronize canvas state across team sessions via git. Ensures all team members see the same product knowledge.