A Claude Skill for autism support is a folder with a SKILL.md file that gives Claude standing instructions for one job, such as drafting a visual schedule or a social story, in literal, low-ambiguity language. Claude loads it automatically when a request matches, across Claude apps, Claude Code, and the API. Building one takes a clear job, a well-written description, and a human reviewing every output.
What Is a Claude Skill?
Anthropic's Agent Skills, announced on October 16, 2025, package instructions, metadata, and optional resources into a folder Claude can load when it is relevant to the request. A Skill is not a one-off prompt. It is saved, reusable expertise that turns a general-purpose assistant into a specialist for a specific, repeatable job, and it loads on demand instead of being retyped into every conversation.
Every Skill needs one required file, SKILL.md, with YAML frontmatter at the top. Two fields are required: name, which must be lowercase letters, numbers, and hyphens only, and description, which states what the Skill does and when Claude should use it. Anthropic's documentation is specific that the description is what Claude matches a request against, so a vague description means a Skill that never fires.
Below the frontmatter, the body of SKILL.md holds the actual instructions: the steps, templates, and guardrails Claude follows once it decides the Skill applies. A Skill can also bundle extra files, additional instruction documents, reference material, and runnable scripts, that Claude only opens when the task actually needs them.
How Claude Loads a Skill
Anthropic's documentation describes this as progressive disclosure, and it happens in three stages. First, the name and description from every installed Skill load at startup, at roughly 100 tokens each, so Claude always knows what is available without any context cost. Second, when a request matches a Skill's description, Claude reads the full SKILL.md body, typically under 5,000 tokens. Third, any bundled reference files or scripts load only if the instructions point Claude to them, and a script's own code never enters the conversation, only its output does.
That design means you can hand Claude many Skills without bloating every conversation. Most of a Skill's content stays off to the side until it is actually needed, the same way a Claude Skills module is built to handle one job well instead of trying to do everything at once.
Where You Can Use a Claude Skill
Custom Skills work the same way wherever you build them, but each surface handles storage differently:
- • Claude apps (claude.ai). Upload a Skill as a zip file in Settings under Features. This is available to Pro, Max, Team, and Enterprise users with code execution turned on, and Claude invokes it automatically once it is installed.
- • Claude Code. Skills are plain folders on disk: ~/.claude/skills/ for a personal Skill you use everywhere, or .claude/skills/ inside a project to share it with that codebase. Claude discovers and loads them with no upload step.
- • The Claude API. Custom Skills are uploaded through the /v1/skills endpoints and are shared workspace-wide, so every member of the workspace can use a Skill one person created.
Custom Skills do not sync between these three surfaces. A Skill built in Claude Code has to be uploaded separately if you also want it on claude.ai or through the API.
Why Build One for Autism Support
An autism-support Skill is useful because the same few requests come up again and again: a visual schedule for a new routine, a social story ahead of an unfamiliar situation, a plan for a sensory break during a long day. Writing the instructions once, in the language an autistic person or their family actually wants, means every future request gets that same careful handling instead of a fresh guess.
The instructions should be autistic-affirming from the first line: identity-first language, acceptance rather than a push toward neurotypical norms, and a clear statement that this tool supports autistic people and families rather than trying to change them. Our roundup of the best AI tools for autistic people and their families covers the broader landscape this kind of Skill fits into, and the same standard applies here: useful AI reduces friction, it does not try to cure anything.
This is not therapy and not medical advice. A Skill like this drafts material a person, parent, or caregiver reviews and adapts, the same way any AI output needs a second look before it goes anywhere that matters.
Building the Skill, Step by Step
Step 1: Name the one job. A Skill that tries to do everything fires unreliably. Pick a specific, repeatable task, such as drafting visual schedules, social stories, and sensory-break plans in a consistent, literal style.
Step 2: Write a description Claude can match against. The description field needs both what the Skill does and when to use it, in plain language, since that is the only text Claude sees before deciding to load the rest. "Drafts visual schedules, social stories, and sensory-break plans for autistic people and their families, using literal, low-ambiguity language" tells Claude exactly when to reach for it.
Step 3: Write the instructions in SKILL.md. This is where the real craft is. Spell out the autistic-affirming framing, the literal communication style, and the format each output type should take, the same discipline that makes a well-built prompt work: a clear role, clear context, and a clear format beat a vague request every time.
Step 4: Add templates if the format is fixed. If visual schedules or social stories should always follow the same shape, a bundled template file keeps every output consistent without retyping the structure in the main instructions.
Step 5: Install it where you will use it. Drop the folder in ~/.claude/skills/ for Claude Code, or zip it and upload it through claude.ai Settings or the API, depending on where the person using it actually works.
Step 6: Review before anything ships. Read every draft before it reaches the person it is for. A Skill makes Claude's output more consistent, not infallible, and nothing it drafts should go out unreviewed, especially material meant for a child or a sensitive moment.
An Example SKILL.md
Here is a compact version covering the four areas above: visual schedules, social stories, sensory breaks, and the literal communication style that ties them together.
---
name: autism-support-drafts
description: Drafts visual schedules, social stories, and sensory-break plans for autistic people and their families. Use when asked for a routine, a social story, a sensory break plan, or literal step-by-step instructions for an autistic person or family.
---
# Autism Support Drafts
## Communication style, always
Write in literal, low-ambiguity language. No idioms, sarcasm, or implied meaning. Use short sentences and concrete nouns. Use identity-first language ("autistic person," not "person with autism," unless the person has said they prefer otherwise). Frame every draft as acceptance and support, never as a way to make someone seem less autistic.
## Visual schedule / routine builder
When asked for a schedule or routine: break it into numbered, concrete steps. One action per step. Note the approximate time or order word (first, next, then, last) instead of assuming a clock is in use. Flag any step that involves a change from the usual routine.
## Social story drafting
When asked for a social story: describe the situation factually in the first person or second person, state what will likely happen, state what the person can do, and avoid assumptions about how the person will feel. Keep sentences short and keep the tone calm and descriptive, not persuasive.
## Sensory-break planning
When asked for a sensory break plan: ask what sensory input is the problem (noise, light, crowding, touch, or unknown) if it is not stated. Offer two or three concrete break options tied to that input. Keep each option specific enough to act on immediately.
## Always close with
A one-line reminder that this draft should be reviewed by a person who knows the individual before it is used, and that it is not therapy or medical advice.A Note on Privacy
Whatever you put into a Skill's instructions or examples can resurface in its output, so keep specific medical details, diagnoses, or other sensitive personal information out of any Skill you might share, upload to a workspace, or hand to someone else. Anthropic's documentation also notes that Agent Skills is not covered by zero data retention arrangements, so Skill content and execution data follow Anthropic's standard retention policy rather than being deleted immediately. If privacy matters for a specific family's situation, keep the details in the conversation itself rather than baked into the Skill file.
Review Every Output
Treat a Skill's draft the way you would treat a first draft from any assistant: read it before anyone relies on it. This is especially true here, since the people reading a social story or using a sensory-break plan may be a child or someone in a stressful moment, and a mismatched detail matters more than it would in a generic business email. The Skill makes Claude consistent at the job. The review is what makes it safe to use.
Sources
Tell us what is costing you the most time. We will map out exactly what your business needs. Free, no obligation.
Build An Agent