Skill Creator

Create new skills for jarvis following the claude code skills standard.

Before Creating a Skill

Always ask the user first. Clarify:

•what the skill should do
•suggested name
•whether it should auto-trigger or be manual-only

Skill Location

All jarvis skills go in: .claude/skills/<skill-name>/SKILL.md

SKILL.md Structure

Every skill needs a SKILL.md with yaml frontmatter:

yaml

---
name: skill-name                        # lowercase, hyphens, max 64 chars
description: When to use this skill...  # include trigger phrases
# optional fields below:
disable-model-invocation: false         # true = only user can invoke via /skill-name
user-invocable: true                    # false = only claude can auto-invoke
argument-hint: "[arg1] [arg2]"          # hint for autocomplete
context: fork                           # run in isolated subagent
agent: Explore                          # subagent type (with context: fork)
---

# Skill Title

Clear instructions for what to do when this skill is invoked.

## Additional Resources

Reference supporting files if needed:
- [references/detailed-docs.md](references/detailed-docs.md)

Key Frontmatter Fields

Field	Use When
`disable-model-invocation: true`	skill has side effects (deploy, send message, commit)
`user-invocable: false`	background knowledge only, not a command
`context: fork`	heavy research task that needs isolated context
`argument-hint`	skill takes arguments like `/skill issue-123`

Note on Permissions

Jarvis runs with bypass permissions, so allowed-tools is only needed if you want to restrict a skill (e.g., read-only skill). Otherwise, omit it.

Directory Structure

For simple skills:

code

skill-name/
└── SKILL.md

For complex skills:

code

skill-name/
├── SKILL.md
├── references/          # detailed docs
│   └── patterns.md
├── examples/            # code examples
│   └── example.py
└── scripts/             # utility scripts
    └── helper.sh

Writing Good Descriptions

Include specific trigger phrases:

yaml

# good
description: Use when asked to "deploy the app", "push to production", or "release"

# bad
description: Handles deployment

Dynamic Context

Skills can inject live data using !command`` syntax:

markdown

Current git status: !`git status --short`

This executes before the skill loads and inserts the output.

Workflow

•Discuss with the user what the skill should do
•Create directory: .claude/skills/<name>/
•Write SKILL.md with frontmatter + instructions
•Add references/ if detailed docs needed
•Test by invoking /skill-name or triggering naturally
•Commit and push changes

Reference

For full frontmatter options and advanced patterns, see references/frontmatter.md