Skills Format (v4.6)

Claude Commands Library now uses the native Claude Code Skills format alongside the original .claude/commands/ format. Both work — Skills add powerful new capabilities.

What Changed

Backward Compatible

Your existing .claude/commands/*.md files work unchanged. Skills are an addition, not a replacement.

Format	Location	Auto-invoke	YAML frontmatter	Supporting files
Commands (old)	.claude/commands/name.md	No	No	No
Skills (new)	.claude/skills/name/SKILL.md	Yes	Yes	Yes

Both create the same /name slash command. Skills have more features.

YAML Frontmatter

Skills use YAML frontmatter to configure behavior:

---
name: prompt
description: Transform any prompt into an unambiguous, executable format.
             Use when the user wants to refine or clarify a task.
argument-hint: "[your prompt or task description]"
disable-model-invocation: false
---

# Skill content starts here...

Frontmatter Fields

Field	Required	Description
name	No	Skill name (defaults to directory name)
description	Recommended	When to use this skill — Claude uses this for auto-invocation
argument-hint	No	Shown in autocomplete (e.g. [issue-number])
disable-model-invocation	No	true = only user can invoke (not Claude)
allowed-tools	No	Tools available without permission prompt
model	No	Override model for this skill
context	No	fork = run in isolated subagent

Auto-Invocation

The description field enables Claude to suggest or auto-invoke a skill when relevant:

description: Transform any prompt into an unambiguous, executable format.
             Use when the user wants to refine or perfect a prompt, clarify
             a vague request, or ensure a task is well-specified.

When you write something like "can you help me make this task description clearer?", Claude sees the description and activates /prompt automatically — no manual invocation needed.

Controlling Who Invokes

# Default: both user and Claude can invoke
disable-model-invocation: false

# Only user can invoke (Claude won't auto-trigger)
disable-model-invocation: true

Use disable-model-invocation: true for workflow commands with side effects:

---
name: reflect
description: Analyze session signals and propose skill file improvements
disable-model-invocation: true  # Don't auto-reflect, user must trigger
---

Project-Aware STARTUP Section

All prompt skills in this library include a STARTUP section that loads project context before any analysis:

## STARTUP: Load Project Context (ALWAYS FIRST)

Before any analysis, load known facts:
1. Read `.claude/memory/project-profile.md`
2. Read last 3 sessions from `.claude/memory/sessions.md`
3. Read `.claude/memory/prompt-patterns.md` if exists

Display:
CONTEXT LOADED FROM PROJECT PROFILE
Project: [name and version]
Stack: [tech stack]
Preferences: [key preferences]
Recent: [one-line summary from sessions.md]

This pattern ensures users never need to repeat project context.

Skills Directory Structure

.claude/
├── commands/              # Old format (still works)
│   ├── prompt.md
│   └── prompt-hybrid.md
└── skills/                # New Skills format
    ├── prompt/
    │   └── SKILL.md       # /prompt with YAML frontmatter
    ├── prompt-hybrid/
    │   └── SKILL.md       # /prompt-hybrid
    ├── prompt-dotnet/
    │   └── SKILL.md       # /prompt-dotnet
    ├── prompt-react/
    │   └── SKILL.md       # /prompt-react
    ├── deploy/
    │   └── SKILL.md       # /deploy
    ├── new-stack/
    │   └── SKILL.md       # /new-stack
    └── reflect/
        └── SKILL.md       # /reflect

Skills in this library are in .claude/skills/. They add YAML frontmatter and STARTUP sections to the existing command logic.

Supporting Files

Skills can include additional files in their directory:

.claude/skills/my-skill/
├── SKILL.md              # Main instructions (required)
├── examples.md           # Usage examples
├── reference.md          # Detailed API reference
└── scripts/
    └── helper.py         # Scripts Claude can run

Reference them from SKILL.md:

For detailed API reference, see [reference.md](reference.md).
For examples, see [examples.md](examples.md).

This keeps SKILL.md focused while letting Claude access details on demand.

Dynamic Context Injection

Skills support !command`` syntax to inject real-time data:

---
name: pr-review
description: Review a pull request
context: fork
allowed-tools: Bash(gh *)
---

## PR Context
- Diff: !`gh pr diff`
- Comments: !`gh pr view --comments`
- Files: !`gh pr diff --name-only`

Review this PR for quality issues...

The shell commands run before Claude sees anything. Output replaces the placeholder.

Subagent Execution

Add context: fork to run a skill in an isolated subagent:

---
name: deep-research
description: Research a topic without polluting main context
context: fork
agent: Explore
---

Research $ARGUMENTS:
1. Find relevant files using Glob and Grep
2. Read and analyze
3. Return summary with file references

This is ideal for codebase exploration — keeps your main conversation clean.

Arguments

Pass arguments to skills with $ARGUMENTS:

---
name: fix-issue
description: Fix a GitHub issue
disable-model-invocation: true
---

Fix GitHub issue $ARGUMENTS following our coding standards.

1. Read the issue: gh issue view $ARGUMENTS
2. Understand the problem
3. Implement the fix
4. Write tests
5. Create a commit

Run: /fix-issue 1234

Positional access: $ARGUMENTS[0], $ARGUMENTS[1], or shorthand $0, $1.

Where Skills Live

Location	Path	Scope
Personal	~/.claude/skills/name/SKILL.md	All your projects
Project	.claude/skills/name/SKILL.md	This project only
Enterprise	Managed settings	All users in org

Migrating from Commands

Commands automatically work as skills — no migration required. To get Skills features:

1. Create .claude/skills/your-command/ directory
2. Add SKILL.md with YAML frontmatter
3. Copy content from your command file
4. Add description for auto-invocation
5. Add STARTUP section for project awareness

Both files can coexist. When names conflict, skills take precedence.

Example: Converting a Command to a Skill

Before (.claude/commands/deploy.md):

# /deploy - Deploy to staging

Run the deployment process:
1. Run tests
2. Build
3. Push to staging

After (.claude/skills/deploy/SKILL.md):

---
name: deploy
description: Deploy the application to staging or production
disable-model-invocation: true
argument-hint: "[staging|production]"
allowed-tools: Bash(git *), Bash(npm *)
---

# /deploy - Deploy to $ARGUMENTS

## STARTUP: Load Project Context

Read `.claude/memory/project-profile.md` for:
- Deployment targets and URLs
- Required pre-deploy checks
- Environment-specific commands

## Deploy Process

1. Run: `npm test` (verify all pass)
2. Build: `npm run build`
3. Push to $ARGUMENTS

...

Related

• Phase 0 Flow - Memory recall and completeness check
• Library System - How commands share core logic
• Commands Overview - All available commands
• Changelog - v4.6 release notes