From amp
PRIMARY skill creator. Use this by default when creating ANY skill. If user explicitly asks for a "Claude skill", "Claude-compatible skill", or "universal skill", use agent-skill-creator instead. Handles Amp-specific features (mcp.json, OAuth, Amp frontmatter).
How this skill is triggered — by the user, by Claude, or both
Slash command
/amp:amp-skill-creatorThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
This skill covers Amp-specific features for skill creation. After reading this, **load the `agent-skill-creator` skill** and follow its workflow, applying the overrides at the end of this document.
This skill covers Amp-specific features for skill creation. After reading this, load the agent-skill-creator skill and follow its workflow, applying the overrides at the end of this document.
Before creating a skill, ask the user where they want it installed (if they haven't already specified):
.agents/skills/ — workspace-local (project-specific)~/.config/amp/skills/ — global userBeyond the required name and description, Amp supports:
---
name: my-skill-name
description: What the skill does and when to use it
argument-hint: "[query]" # Shown in /skill-list (e.g., "[repo] [issue]")
disable-model-invocation: true # Hides from agent auto-detection (manual /skill only)
---
Skills can bundle MCP servers via mcp.json in the skill root:
skill-name/
├── SKILL.md
└── mcp.json
{
"local-server-name": {
"command": "npx",
"args": ["-y", "some-mcp-server@latest"],
"env": {
"API_KEY": "${MY_API_KEY}"
},
"includeTools": ["tool_a", "tool_b", "navigate_*"]
},
"remote-server-name": {
"url": "https://some-mcp-server.com/mcp",
"headers": {
"Authorization": "Bearer ${TOKEN}"
},
"includeTools": ["tool_a", "tool_b"]
}
}
| Field | Required | Description |
|---|---|---|
command | Yes | Command to run |
args | No | Array of command arguments |
env | No | Environment variables for the server |
includeTools | Recommended | Glob patterns to filter exposed tools |
| Field | Required | Description |
|---|---|---|
url | Yes | URL of the MCP server |
headers | No | Headers to send with requests |
includeTools | Recommended | Glob patterns to filter exposed tools |
Use ${VAR_NAME} syntax in any field:
{
"my-server": {
"command": "node",
"args": ["${HOME}/scripts/mcp-server.js"],
"env": {
"API_KEY": "${SERVICE_API_KEY}"
}
}
}
MCP servers often expose many tools (20+ = thousands of tokens). Always use includeTools:
{
"includeTools": ["navigate_page", "take_screenshot", "click"]
}
Glob patterns supported: ["navigate_*", "click"] or ["*"] for all.
Use amp mcp add to quickly generate MCP server config:
# Local server (command-based)
amp mcp add <name> -- <command> [args...]
# Local server with env vars
amp mcp add <name> --env KEY=VAL -- <command> [args...]
# Remote server (URL-based, auto-detects transport)
amp mcp add <name> <url>
# Remote server with headers
amp mcp add <name> --header "Authorization=Bearer <token>" <url>
# Add to workspace settings instead of global
amp mcp add <name> --workspace -- <command> [args...]
Note: This command adds the server config to
~/.config/amp/settings.json(or workspace settings with--workspace), not directly into a skill'smcp.json. After running, copy the relevant entry from settings into your skill'smcp.jsonfile.
| Option | Description |
|---|---|
--env KEY=VAL | Environment variables (repeatable) |
--header KEY=VAL | HTTP headers for URL-based servers (repeatable) |
--workspace | Add to workspace settings instead of global (~/.config/amp/settings.json) |
# NPX-based server
amp mcp add context7 -- npx -y @upstash/context7-mcp
# Postgres with env vars
amp mcp add postgres --env PGUSER=myuser -- npx -y @modelcontextprotocol/server-postgres postgresql://localhost/mydb
# Remote with auth header
amp mcp add sourcegraph --header "Authorization=token <token>" https://sourcegraph.example.com/.api/mcp/v1
# Remote with OAuth (auto-triggers browser auth)
amp mcp add linear https://mcp.linear.app/sse
# Workspace-specific server
amp mcp add project-server --workspace -- npx -y @some/server
Load the agent-skill-creator skill now and follow its 6-phase workflow for creating the skill content. Apply these overrides during implementation:
| agent-skill-creator Says | Amp Does Instead |
|---|---|
Create marketplace.json | Not used — Amp uses SKILL.md frontmatter only |
Create .claude-plugin/ directory | Not used in Amp |
Skills in plugins/cache/ | Use locations from "First: Ask Where to Install" above |
plugins[0].description must sync with SKILL.md | Not applicable — only SKILL.md frontmatter matters |
| Cross-platform export to Desktop/API | Not applicable to Amp skills |
Use -cskill suffix naming convention | Optional — use descriptive names |
Run /plugin marketplace add ./skill-name | Skill will be loaded automatically based on file location |
What to USE from agent-skill-creator:
scripts/, utils/)references/)Provides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
npx claudepluginhub thurstonsand/ansiblonomicon --plugin amp