Mcp generate instructions
Skill dequelabs/axe-accessibility/skills/mcp-generate-instructions
This skill should be used when the user asks to "generate accessibility instructions", "add axe instructions to my repo", "bake a11y into my coding agent", "create copilot-instructions for accessibility", "set up CLAUDE.md for axe", or runs /axe-accessibility:mcp-generate-instructions. It writes or updates agent-instruction files (CLAUDE.md, .github/copilot-instructions.md, Cursor rules, or AGENTS.md) that enforce the Axe MCP analyze/igt -> remediate -> verify workflow.From its SKILL.md
npx -y skills add dequelabs/axe-accessibility --skill mcp-generate-instructionsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
5.0 KB, as published. Nobody here has run it
Generate accessibility agent instructions
Produce agent-instruction files that make a coding agent automatically run Deque's Axe MCP analyze -> remediate -> verify workflow whenever it touches UI. The goal is the same effect as a hand-written copilot-instructions.md: every UI change is checked and fixed against the axe engine before it is considered done.
Step 1 — Determine targets
Targets and their files:
| Target | File |
|---|---|
claude | CLAUDE.md (repo root) |
copilot | .github/copilot-instructions.md |
cursor | .cursor/rules/accessibility.mdc (fallback .cursorrules) |
agents | AGENTS.md (repo root) |
If $1 specifies targets (claude, copilot, cursor, agents, or all), use them. Otherwise detect what already exists in the repo with Glob (CLAUDE.md, .github/copilot-instructions.md, .cursor/**, .cursorrules, AGENTS.md) and confirm the set with AskUserQuestion — default to the files already present, offering to add others.
Step 2 — Compose the content
Use the canonical workflow text in references/workflow-template.md as the body. Adapt the tool-name references per target — see references/targets.md for each target's tool-naming convention and any file-format specifics (e.g. Cursor .mdc frontmatter).
Keep these intact across all targets — they are the parts that change agent behavior:
- the mandatory analyze -> remediate -> verify loop
- the
remediatebatching rule (one call per scan,idon every issue) — the most common cause of failed tool calls - the field mapping, including reading issues from
response.dataand never passing an issue's ownremediationobject into theremediationparameter - the deterministic (
isAdvanced: false) vs. AI-derived (isAdvanced: true) trust distinction - the
igtkeyboard section and its different issue shape - the credit-aware note and the image-alt guidance
Step 2a — Refresh outdated sections
When updating a file that already has an axe workflow section, check for guidance written against the pre-1.3.0 tool contract and correct it rather than appending alongside it. Tell-tale signs:
- a per-issue
remediatecall, or "callremediateonce per violation" - a flat
remediate({pageUrl, rule, elementHtml, remediation})signature with noidand noissuesarray - reading issues from
response.issuesinstead ofresponse.data remediationmapped from onlydescription+helpText, with no mention ofsummary- no mention of
igt - a blanket claim that all findings are deterministic / zero-false-positive
Replace those in place — leaving them causes calls the server will reject.
Step 3 — Write without clobbering
For each target file:
- If the file does not exist, create it with the composed content.
- If it exists, merge rather than overwrite. Read it first. If it already contains an axe/accessibility workflow section (look for a heading like "Accessibility Testing and Remediation Workflow"), update that section in place. Otherwise append the section under a clear heading, preserving all existing content.
- Never delete unrelated instructions the user already maintains.
Show the user a brief diff/summary of what was added or changed per file.
Step 4 — Tailor to the repo (optional but recommended)
Inspect the repo to make the instructions concrete:
- Detect the dev-server URL and start command (check
package.jsonscripts,README, framework config) and reference the real localhost URL in the analysis step. - If the repo has project-specific asset conventions worth encoding (e.g. SVGs whose meaningful parts are identified by comments), add a short project-specific note mirroring the pattern in
references/workflow-template.md. Ask the user before inventing project-specific rules.
Step 5 — Close out
Summarize which files were written/updated and remind the user that:
- The instructions only take effect for agents that read them (Copilot reads
.github/copilot-instructions.md, Claude Code readsCLAUDE.md, etc.). - The Axe MCP Server must be connected (
/axe-accessibility:mcp-setup) for the workflow to function. /axe-accessibility:mcp-audit <url>runs the same loop on demand.
Additional resources
references/workflow-template.md— the canonical mandatory-workflow text to embed.references/targets.md— per-target file paths, tool-naming conventions, and format notes.