From ac
Create Claude Code skills with progressive disclosure architecture. Use when building reusable workflow skills, domain-specific skills, or optimizing existing skills for Claude Code plugins.
npx claudepluginhub anilcancakir/claude-code-plugin --plugin acThis skill uses the workspace's default tool permissions.
Create high-quality Claude Code skills following progressive disclosure — metadata always loaded, SKILL.md body on trigger, references on demand.
Generates design tokens/docs from CSS/Tailwind/styled-components codebases, audits visual consistency across 10 dimensions, detects AI slop in UI.
Records polished WebM UI demo videos of web apps using Playwright with cursor overlay, natural pacing, and three-phase scripting. Activates for demo, walkthrough, screen recording, or tutorial requests.
Delivers idiomatic Kotlin patterns for null safety, immutability, sealed classes, coroutines, Flows, extensions, DSL builders, and Gradle DSL. Use when writing, reviewing, refactoring, or designing Kotlin code.
Create high-quality Claude Code skills following progressive disclosure — metadata always loaded, SKILL.md body on trigger, references on demand.
1. Interview
Gather requirements through focused questions. Limit to 3-4 AskUserQuestion calls maximum:
Skip questions where the answer is obvious from context.
2. Explore Existing Patterns
Launch 1-2 ac:explore agents to find existing conventions in the target project:
3. Dedup Audit
Read ${CLAUDE_SKILL_DIR}/../prompt-writer/references/cc-dedup-guide.md before drafting. Check what CC already provides natively — the skill body must not repeat general concepts the model already knows. Cut anything CC handles by default.
4. Draft Skill
Follow progressive disclosure. SKILL.md body ≤500 lines. Move detail to references/. Read references/skill-patterns.md for description patterns, invocation control, string substitutions, and directory structure before writing.
Key decisions at this stage:
user-invocable: false? disable-model-invocation: true? (see Invocation Control Matrix)context: fork if the skill is self-contained and parallelizable?paths: if the skill activates only for specific file types?5. Wire References
CC does NOT auto-load references/ files. The model never sees them unless SKILL.md points to them explicitly.
## References table at the end for 3+ files.references/, scripts/, assets/ is reachable — orphan files are invisible.6. Frontmatter Validation
Read ${CLAUDE_SKILL_DIR}/../prompt-writer/references/frontmatter-schemas.md for the complete skill frontmatter schema. Key fields: name, description, when_to_use, user-invocable, disable-model-invocation, model, effort, allowed-tools, context, agent, argument-hint, arguments, paths, hooks.
7. Review
Present the draft. Run the quality checklist below before delivering.
Design every skill with three loading levels.
Level 1 — Metadata (~100 tokens)
name + description injected into <system-reminder> listingLevel 2 — SKILL.md body (≤5k tokens)
/nameLevel 3 — Bundled resources (unlimited)
references/ — detailed docs, API refs, schemasscripts/ — executable helpers (stdout consumed, source NOT loaded)assets/ — HTML templates, static filesSKILL.md → reference.md works. Deeper chains breakscripts/ — executed, not loaded (only stdout consumes tokens)references/Run before presenting any skill to the user.
Frontmatter
when_to_use present for auto-triggered skills?user-invocable / disable-model-invocation set correctly?allowed-tools minimal — only what the skill actually needs?paths field used if file-type specific? context: fork if parallelizable?Content 6. Every sentence is an action? ("Consider" → "Use", "Extract", "Run") 7. No unnecessary concept explanations? (the model already knows them) 8. No vague words? ("good", "proper" → concrete rule) 9. No repetition across sections? 10. CRITICAL/IMPORTANT inflation? Maximum 1-2 per file
Progressive Disclosure
11. SKILL.md body <500 lines? References >100 lines have TOC?
12. Every reference explicitly linked with "read this when" guidance?
13. All files in references/ reachable from SKILL.md?
14. One level deep only — no chained reference files?
Style 15. Imperative mood throughout? No passive voice? 16. Nesting depth ≤1? Sentences under 25 words? 17. No hardcoded paths or usernames?
| Topic | File | When to read |
|---|---|---|
| Skill patterns | skill-patterns.md | Before drafting — description patterns, invocation matrix, string substitutions, directory structure |
| Shared writing conventions | ${CLAUDE_SKILL_DIR}/../prompt-writer/references/cc-dedup-guide.md | Before drafting — what CC already provides natively (skip in skill body) |
| Frontmatter schema | ${CLAUDE_SKILL_DIR}/../prompt-writer/references/frontmatter-schemas.md | During frontmatter validation — complete field reference |