[IMPORTANT] Use TaskCreate to break ALL work into small tasks BEFORE starting.
Shared Protocol Duplication Policy — Inline protocol content in skills (wrapped in <!-- SYNC:tag -->) is INTENTIONAL duplication. Do NOT extract, deduplicate, or replace with file references. AI compliance drops significantly when protocols are behind file-read indirection. To update: edit .claude/skills/shared/sync-inline-versions.md first, then grep SYNC:protocol-name and update all occurrences.
Output Quality — Token efficiency without sacrificing quality.
- No inventories/counts — AI can
grep | wc -l. Counts go stale instantly
- No directory trees — AI can
glob/ls. Use 1-line path conventions
- No TOCs — AI reads linearly. TOC wastes tokens
- No examples that repeat what rules say — one example only if non-obvious
- Lead with answer, not reasoning. Skip filler words and preamble
- Sacrifice grammar for concision in reports
- Unresolved questions at end, if any
Quick Summary
Goal: Create new Claude Code skills with proper structure or scan/fix invalid skill headers across the catalog.
Workflow:
- Clarify — Gather purpose, trigger keywords, tools needed via AskUserQuestion
- Check Existing — Glob for similar skills, avoid duplication
- Scaffold — Create directory + SKILL.md with frontmatter + Quick Summary
- Add SYNC blocks — Include relevant protocol checklists from
sync-inline-versions.md
- Validate — Run frontmatter + header validation
- Enhance — Call
/prompt-enhance on the new SKILL.md for attention anchoring quality
Key Rules:
- Every SKILL.md MUST ATTENTION include
## Quick Summary (Goal/Workflow/Key Rules) within first 30 lines
- Single-line
description with [Category] prefix and trigger keywords
- SKILL.md under 500 lines; use
references/ for detail
- Shared protocols MUST ATTENTION be inlined via
<!-- SYNC:tag --> blocks, NEVER file references
- MUST ATTENTION call
/prompt-enhance on new/updated skills as final quality pass
Modes
| Mode |
Trigger |
Action |
| Create |
$ARGUMENTS describes a new skill |
Create skill following workflow below |
| Scan & Fix |
$ARGUMENTS mentions scan, fix, validate, invalid |
Run validation across all skills |
SKILL.md Schema Reference
Frontmatter Fields (all optional)
---
name: my-skill # Lowercase, hyphens. Max 64 chars. Default: directory name.
description: '[Category] What it does. Triggers on: keyword1, keyword2.' # MUST ATTENTION be single-line
argument-hint: '[issue-number]' # Autocomplete hint for arguments
disable-model-invocation: false # true = user-only (/name), Claude cannot auto-invoke
user-invocable: true # false = hidden from / menu, Claude-only auto-invoke
context: inline # inline (default) or fork (isolated subagent)
agent: general-purpose # Subagent type when context: fork
model: opus-4-5 # Model override. Default: session model
allowed-tools: 'Read, Grep, Glob' # Comma-separated tool restrictions
version: 1.0.0 # Project convention (non-official)
---
Invocation Control Matrix
| Setting |
User Invokes |
Claude Invokes |
Description |
| Default |
Yes |
Yes |
Description in context; loads on invocation |
disable-model-invocation: true |
Yes |
No |
User-only. Not in context until invoked |
user-invocable: false |
No |
Yes |
Hidden from menu. Claude auto-invokes |
Variable Substitution
| Variable |
Description |
$ARGUMENTS |
All arguments passed to skill |
$ARGUMENTS[N] / $N |
Specific argument by 0-based index |
${CLAUDE_SESSION_ID} |
Current session ID |
!`command` |
Execute shell command before skill runs |
Key Rules
- Single-line
description (multi-line YAML breaks catalog parsing)
- Include trigger keywords in description for auto-activation
- Use
[Category] prefix (e.g., [Frontend], [Planning], [AI & Tools])
- SKILL.md under 500 lines; use
references/ for detail
- Descriptions loaded at 2% of context window (~16,000 chars)
- Include "ultrathink" in skill content to enable extended thinking mode
Mode 1: Create Skill
Workflow
- Clarify — If requirements unclear, use
AskUserQuestion for: purpose, auto vs user-invoked, trigger keywords, tools needed
- Check Existing — Glob
.claude/skills/*/SKILL.md for similar skills. Avoid duplication.
- Create Directory —
.claude/skills/{skill-name}/SKILL.md
- Write Frontmatter — Follow schema from
## SKILL.md Schema Reference section above
- Write Instructions — Concise, actionable, progressive disclosure
- Add SYNC Blocks — Identify which shared protocols apply to this skill. Read
.claude/skills/shared/sync-inline-versions.md and copy relevant <!-- SYNC:tag --> blocks inline. Common protocols: understand-code-first, evidence-based-reasoning, output-quality-principles
- Add Closing Reminders — Echo top rules at bottom with
:reminder SYNC blocks for recency anchoring
- Add References — Move detailed docs to
references/ directory if content >200 lines
- Add Scripts — Create
scripts/ for executable helpers if needed
- Validate — Run frontmatter validation (see Mode 2 single-file check)
- Enhance — Call
/prompt-enhance on the finished SKILL.md for AI attention anchoring quality
Frontmatter Template
---
name: { kebab-case-name }
description: '[Category] What it does. Triggers on: keyword1, keyword2.'
---
Skill Attention Structure (MUST ATTENTION follow)
[Frontmatter]
[SYNC protocol blocks — top attention zone]
[## Quick Summary — Goal/Workflow/Key Rules]
[Detailed instructions — middle zone]
[## Closing Reminders — bottom attention zone with :reminder SYNC blocks]
Why: AI attention is strongest at TOP and BOTTOM (primacy-recency effect). Place critical rules in both zones. See /prompt-enhance for research-backed principles.
SYNC Tag Protocol for New Skills
- Read
.claude/skills/shared/sync-inline-versions.md — canonical source for all protocol checklists
- Identify which protocols the skill needs (e.g., investigation skills need
understand-code-first + evidence-based-reasoning)
- Copy the checklist content between
<!-- SYNC:tag --> open/close tags at the TOP of the skill
- Add 1-line
:reminder versions at the BOTTOM inside Closing Reminders
- NEVER use
MUST ATTENTION READ shared/ file references — always inline
Rules
- SKILL.md is instructions, not documentation. Teach Claude HOW to do the task.
- Single-line
description (multi-line YAML breaks catalog parsing)
- Description must include trigger keywords for auto-activation
- Use
[Category] prefix in description (e.g., [Frontend], [Planning], [AI & Tools])
- Keep SKILL.md under 500 lines; use
references/ for detail
- Progressive disclosure: frontmatter → SKILL.md summary → reference files
- Token efficiency: every line must earn its place
Mode 2: Scan & Fix Invalid Skills
What It Validates
| Check |
Rule |
Severity |
| Frontmatter exists |
Must have --- delimiters |
Error |
| Description single-line |
No literal newlines in description value |
Error |
| Description not empty |
Must have description for discoverability |
Warning |
| Name format |
Lowercase, hyphens, max 64 chars |
Error |
| No unknown official fields |
Flag fields not in official schema |
Info |
| Description has category |
Should start with [Category] |
Warning |
| File size |
SKILL.md should be <500 lines |
Warning |
| Quick Summary exists |
Must have ## Quick Summary in first 30 lines |
Warning |
| SYNC tag balance |
Every open tag must have matching close tag |
Error |
Scan Workflow
- Discover — Glob
.claude/skills/*/SKILL.md for all skills
- Parse — Read first 20 lines of each file, extract frontmatter
- Validate — Check each rule above
- Report — List issues grouped by severity (Error > Warning > Info)
- Fix — If user confirms, fix Error-level issues automatically
Validate Script
node .claude/skills/skill-create/scripts/validate-skills.cjs # Report only
node .claude/skills/skill-create/scripts/validate-skills.cjs --fix # Report + auto-fix
Requirements
$ARGUMENTS
Closing Reminders
- IMPORTANT MUST ATTENTION break work into small todo tasks using
TaskCreate BEFORE starting
- IMPORTANT MUST ATTENTION inline shared protocols via
<!-- SYNC:tag --> blocks — NEVER use MUST ATTENTION READ shared/ file references
- IMPORTANT MUST ATTENTION call
/prompt-enhance on new/updated skills as final attention-anchoring quality pass
- IMPORTANT MUST ATTENTION include
## Quick Summary within first 30 lines of every SKILL.md
- IMPORTANT MUST ATTENTION add Closing Reminders section with
:reminder SYNC blocks at bottom of every skill
- IMPORTANT MUST ATTENTION follow SKILL.md Schema Reference (inlined above) for official frontmatter fields
- IMPORTANT MUST ATTENTION follow duplication policy: inline protocols are INTENTIONAL, never extract to file references
- IMPORTANT MUST ATTENTION follow output quality principles: token efficiency, lead with answer, no filler
Converted and distributed by TomeVault — claim your Tome and manage your conversions.
1---2name: duc01226-easyplatform-skill-create3description: > **[IMPORTANT]** Use `TaskCreate` to break ALL work into small tasks BEFORE starting.4---56> **[IMPORTANT]** Use `TaskCreate` to break ALL work into small tasks BEFORE starting.78<!-- SYNC:shared-protocol-duplication-policy -->910> **Shared Protocol Duplication Policy** — Inline protocol content in skills (wrapped in `<!-- SYNC:tag -->`) is INTENTIONAL duplication. Do NOT extract, deduplicate, or replace with file references. AI compliance drops significantly when protocols are behind file-read indirection. To update: edit `.claude/skills/shared/sync-inline-versions.md` first, then grep `SYNC:protocol-name` and update all occurrences.1112<!-- /SYNC:shared-protocol-duplication-policy -->1314<!-- SYNC:output-quality-principles -->1516> **Output Quality** — Token efficiency without sacrificing quality.17>18> 1. No inventories/counts — AI can `grep | wc -l`. Counts go stale instantly19> 2. No directory trees — AI can `glob`/`ls`. Use 1-line path conventions20> 3. No TOCs — AI reads linearly. TOC wastes tokens21> 4. No examples that repeat what rules say — one example only if non-obvious22> 5. Lead with answer, not reasoning. Skip filler words and preamble23> 6. Sacrifice grammar for concision in reports24> 7. Unresolved questions at end, if any2526<!-- /SYNC:output-quality-principles -->2728## Quick Summary2930**Goal:** Create new Claude Code skills with proper structure or scan/fix invalid skill headers across the catalog.3132**Workflow:**33341. **Clarify** — Gather purpose, trigger keywords, tools needed via AskUserQuestion352. **Check Existing** — Glob for similar skills, avoid duplication363. **Scaffold** — Create directory + SKILL.md with frontmatter + Quick Summary374. **Add SYNC blocks** — Include relevant protocol checklists from `sync-inline-versions.md`385. **Validate** — Run frontmatter + header validation396. **Enhance** — Call `/prompt-enhance` on the new SKILL.md for attention anchoring quality4041**Key Rules:**4243- Every SKILL.md MUST ATTENTION include `## Quick Summary` (Goal/Workflow/Key Rules) within first 30 lines44- Single-line `description` with `[Category]` prefix and trigger keywords45- SKILL.md under 500 lines; use `references/` for detail46- Shared protocols MUST ATTENTION be inlined via `<!-- SYNC:tag -->` blocks, NEVER file references47- MUST ATTENTION call `/prompt-enhance` on new/updated skills as final quality pass4849## Modes5051| Mode | Trigger | Action |52| -------------- | -------------------------------------------------- | ------------------------------------- |53| **Create** | `$ARGUMENTS` describes a new skill | Create skill following workflow below |54| **Scan & Fix** | `$ARGUMENTS` mentions scan, fix, validate, invalid | Run validation across all skills |5556## SKILL.md Schema Reference5758### Frontmatter Fields (all optional)5960```yaml61---62name: my-skill # Lowercase, hyphens. Max 64 chars. Default: directory name.63description: '[Category] What it does. Triggers on: keyword1, keyword2.' # MUST ATTENTION be single-line64argument-hint: '[issue-number]' # Autocomplete hint for arguments65disable-model-invocation: false # true = user-only (/name), Claude cannot auto-invoke66user-invocable: true # false = hidden from / menu, Claude-only auto-invoke67context: inline # inline (default) or fork (isolated subagent)68agent: general-purpose # Subagent type when context: fork69model: opus-4-5 # Model override. Default: session model70allowed-tools: 'Read, Grep, Glob' # Comma-separated tool restrictions71version: 1.0.0 # Project convention (non-official)72---73```7475### Invocation Control Matrix7677| Setting | User Invokes | Claude Invokes | Description |78| -------------------------------- | ------------ | -------------- | ------------------------------------------- |79| Default | Yes | Yes | Description in context; loads on invocation |80| `disable-model-invocation: true` | Yes | No | User-only. Not in context until invoked |81| `user-invocable: false` | No | Yes | Hidden from menu. Claude auto-invokes |8283### Variable Substitution8485| Variable | Description |86| ---------------------- | --------------------------------------- |87| `$ARGUMENTS` | All arguments passed to skill |88| `$ARGUMENTS[N]` / `$N` | Specific argument by 0-based index |89| `${CLAUDE_SESSION_ID}` | Current session ID |90| `` !`command` `` | Execute shell command before skill runs |9192### Key Rules9394- Single-line `description` (multi-line YAML breaks catalog parsing)95- Include trigger keywords in description for auto-activation96- Use `[Category]` prefix (e.g., `[Frontend]`, `[Planning]`, `[AI & Tools]`)97- SKILL.md under 500 lines; use `references/` for detail98- Descriptions loaded at 2% of context window (~16,000 chars)99- Include "ultrathink" in skill content to enable extended thinking mode100101## Mode 1: Create Skill102103### Workflow1041051. **Clarify** — If requirements unclear, use `AskUserQuestion` for: purpose, auto vs user-invoked, trigger keywords, tools needed1062. **Check Existing** — Glob `.claude/skills/*/SKILL.md` for similar skills. Avoid duplication.1073. **Create Directory** — `.claude/skills/{skill-name}/SKILL.md`1084. **Write Frontmatter** — Follow schema from `## SKILL.md Schema Reference` section above1095. **Write Instructions** — Concise, actionable, progressive disclosure1106. **Add SYNC Blocks** — Identify which shared protocols apply to this skill. Read `.claude/skills/shared/sync-inline-versions.md` and copy relevant `<!-- SYNC:tag -->` blocks inline. Common protocols: `understand-code-first`, `evidence-based-reasoning`, `output-quality-principles`1117. **Add Closing Reminders** — Echo top rules at bottom with `:reminder` SYNC blocks for recency anchoring1128. **Add References** — Move detailed docs to `references/` directory if content >200 lines1139. **Add Scripts** — Create `scripts/` for executable helpers if needed11410. **Validate** — Run frontmatter validation (see Mode 2 single-file check)11511. **Enhance** — Call `/prompt-enhance` on the finished SKILL.md for AI attention anchoring quality116117### Frontmatter Template118119```yaml120---121name: { kebab-case-name }122description: '[Category] What it does. Triggers on: keyword1, keyword2.'123---124```125126### Skill Attention Structure (MUST ATTENTION follow)127128```129[Frontmatter]130[SYNC protocol blocks — top attention zone]131[## Quick Summary — Goal/Workflow/Key Rules]132[Detailed instructions — middle zone]133[## Closing Reminders — bottom attention zone with :reminder SYNC blocks]134```135136**Why:** AI attention is strongest at TOP and BOTTOM (primacy-recency effect). Place critical rules in both zones. See `/prompt-enhance` for research-backed principles.137138### SYNC Tag Protocol for New Skills1391401. Read `.claude/skills/shared/sync-inline-versions.md` — canonical source for all protocol checklists1412. Identify which protocols the skill needs (e.g., investigation skills need `understand-code-first` + `evidence-based-reasoning`)1423. Copy the checklist content between `<!-- SYNC:tag -->` open/close tags at the TOP of the skill1434. Add 1-line `:reminder` versions at the BOTTOM inside Closing Reminders1445. NEVER use `MUST ATTENTION READ shared/` file references — always inline145146### Rules147148- SKILL.md is instructions, not documentation. Teach Claude HOW to do the task.149- Single-line `description` (multi-line YAML breaks catalog parsing)150- Description must include trigger keywords for auto-activation151- Use `[Category]` prefix in description (e.g., `[Frontend]`, `[Planning]`, `[AI & Tools]`)152- Keep SKILL.md under 500 lines; use `references/` for detail153- Progressive disclosure: frontmatter → SKILL.md summary → reference files154- Token efficiency: every line must earn its place155156## Mode 2: Scan & Fix Invalid Skills157158### What It Validates159160| Check | Rule | Severity |161| -------------------------- | ---------------------------------------------- | -------- |162| Frontmatter exists | Must have `---` delimiters | Error |163| Description single-line | No literal newlines in description value | Error |164| Description not empty | Must have description for discoverability | Warning |165| Name format | Lowercase, hyphens, max 64 chars | Error |166| No unknown official fields | Flag fields not in official schema | Info |167| Description has category | Should start with `[Category]` | Warning |168| File size | SKILL.md should be <500 lines | Warning |169| Quick Summary exists | Must have `## Quick Summary` in first 30 lines | Warning |170| SYNC tag balance | Every open tag must have matching close tag | Error |171172### Scan Workflow1731741. **Discover** — Glob `.claude/skills/*/SKILL.md` for all skills1752. **Parse** — Read first 20 lines of each file, extract frontmatter1763. **Validate** — Check each rule above1774. **Report** — List issues grouped by severity (Error > Warning > Info)1785. **Fix** — If user confirms, fix Error-level issues automatically179180### Validate Script181182```bash183node .claude/skills/skill-create/scripts/validate-skills.cjs # Report only184node .claude/skills/skill-create/scripts/validate-skills.cjs --fix # Report + auto-fix185```186187## Requirements188189<user-prompt>$ARGUMENTS</user-prompt>190191---192193## Closing Reminders194195- **IMPORTANT MUST ATTENTION** break work into small todo tasks using `TaskCreate` BEFORE starting196- **IMPORTANT MUST ATTENTION** inline shared protocols via `<!-- SYNC:tag -->` blocks — NEVER use `MUST ATTENTION READ shared/` file references197- **IMPORTANT MUST ATTENTION** call `/prompt-enhance` on new/updated skills as final attention-anchoring quality pass198- **IMPORTANT MUST ATTENTION** include `## Quick Summary` within first 30 lines of every SKILL.md199- **IMPORTANT MUST ATTENTION** add Closing Reminders section with `:reminder` SYNC blocks at bottom of every skill200- **IMPORTANT MUST ATTENTION** follow SKILL.md Schema Reference (inlined above) for official frontmatter fields201 <!-- SYNC:shared-protocol-duplication-policy:reminder -->202- **IMPORTANT MUST ATTENTION** follow duplication policy: inline protocols are INTENTIONAL, never extract to file references203 <!-- /SYNC:shared-protocol-duplication-policy:reminder -->204 <!-- SYNC:output-quality-principles:reminder -->205- **IMPORTANT MUST ATTENTION** follow output quality principles: token efficiency, lead with answer, no filler206 <!-- /SYNC:output-quality-principles:reminder -->207208---209> Converted and distributed by [TomeVault](https://tomevault.io/claim/duc01226) — claim your Tome and manage your conversions.210<!-- tomevault:4.0:skill_md:2026-04-13 -->