Designing Skills for Claude Code

Use this skill when creating new skills to ensure they are discoverable and useful.

Critical Insight

The name and description determine whether an agent will EVER use your skill. Get these wrong and the skill will sit unused while the agent struggles without it.

Skill Structure

skills/
└── skill-name/             # lowercase-with-hyphens, matches name in frontmatter
    ├── SKILL.md            # The skill definition (mandatory)
    ├── skill.json          # Optional: declares commands for the runner (see below)
    └── scripts/            # Executable scripts (one per skill command)
        └── my-script.ts    # Convention: scripts/<command-name>.{js,ts,py,sh}

Running Skills via the Anima Skill Runner

All skill scripts are invoked through the Anima runner, which resolves the script, sets cwd, and injects env vars. Scripts no longer need a cd dance:

anima skill run <skill-id> <command> [args...]      # synchronous
anima skill run <skill-id> <command> [...] --task   # queue via scheduler
anima skill task <task-id> --watch                  # poll a queued task
anima skill list                                    # discovery
anima skill help <skill-id> <command>               # per-command help

The runner injects these env vars before exec:

• SKILL_DIR — Absolute path to the skill directory. Use this for any

skill-internal resource. Never use process.cwd().

• SKILL_ID — Skill identifier
• SKILL_COMMAND — Command name
• ANIMATASKID — Only set in --task mode. Pass to

anima scheduler updateprogress --taskId "$ANIMATASK_ID" --message "..." for live progress reporting.

• ANIMAEXECUTIONID — Only set in --task mode

skill.json — Optional Metadata

When absent, any executable file under scripts/ is callable by basename (e.g., scripts/foo.js → anima skill run myskill foo). Runtime is auto-detected by extension or shebang.

For richer behavior — argv help, longRunning auto-queue, required env checks — add a skill.json. Each command resolves to one of two execution modes (mutually exclusive):

script — A file inside the skill directory

For logic that lives with the skill. Relative path. Runtime auto-detected.

{
  "id": "writing-romance-novels",
  "description": "Generate full-length romance novels — chapter MP3s and AI cover art.",
  "commands": {
    "generate-cover": {
      "script": "scripts/generate-cover.js",
      "runtime": "node",
      "longRunning": false,
      "description": "Generate cover art using Gemini Imagen from cover.md inside a novel folder.",
      "args": [{ "name": "novel-folder", "type": "absolute-folder", "required": true }],
      "env": ["GEMINI_API_KEY"],
      "timeoutMs": 180000
    }
  }
}

command — A binary on PATH

For shared tooling. Avoids duplicating logic across multiple skills. The runner spawns the binary directly with full env injection (SKILLDIR, ANIMATASK_ID, etc.) just like script mode.

{
  "id": "writing-romance-novels",
  "commands": {
    "generate-audio": {
      "command": "eleven-tts",
      "longRunning": true,
      "description": "Generate MP3 audio from chapter markdown using the shared eleven-tts binary.",
      "args": [{ "name": "chapter-path", "type": "absolute-file", "required": true }],
      "env": ["ELEVENLABS_API_KEY", "ELEVENLABS_VOICE_ID"],
      "timeoutMs": 1800000
    }
  }
}

When to extract a command: If two or more skills would have an identical (or near-identical) script, extract it once into anima/scripts/<tool>.js, symlink to ~/.local/bin/<tool>, and point each skill's command field at the binary name. Future skills get the tool for free without copy-pasting code.

Effects of longRunning: true: The runner auto-enables --task mode (returns task ID immediately, executes via the scheduler). Pass --sync to override and run inline. Works the same in both script and command modes.

Frontmatter Format

---
name: skill-name # lowercase-with-hyphens
description: "..." # What it does AND when to use it - MUST be quoted
---

Description Requirements

The description MUST include:

1. What the skill does (capabilities)
2. ALL activities covered — If the skill handles creating, reviewing, updating, and auditing, say so explicitly. Users phrase requests differently ("assess", "check", "audit", "review", "modify", "update", "create", "build"). Include all that should trigger your skill.
3. When to use it (trigger conditions)
4. Action words like "MUST be loaded before..." or "Use PROACTIVELY when..." for skills that should auto-invoke

Naming Conventions

• Use gerund form (verb ending in -ing): developing-, processing-, managing-, setting-up-
• Use plural nouns: developing-skills, processing-images, managing-ads

Example: Good Description

name: browsing-the-web
description: "MUST be used when you need to browse the web. Efficient browser automation designed for agents - enables intuitive web navigation, form filling, screenshots, and data scraping through accessibility-based workflows. Triggers on: browse website, visit URL, open webpage, fill form, click button, take screenshot, scrape data, web automation, interact with website."

Why it works:

• Starts with "MUST be used when..." (strong trigger)
• Lists capabilities (navigation, forms, screenshots, scraping)
• Includes many trigger phrases users might say

Example: Transcription Skill

name: transcribing-audio
description: "MUST be used when you need to transcribe audio files to text. Local speech-to-text (STT) transcription using Parakeet MLX on Apple Silicon - fast, private, offline. Triggers on: transcribe audio, convert audio to text, speech to text, STT, transcription, get text from audio, audio file transcription, voice to text, extract text from recording, transcribe podcast, transcribe meeting, transcribe voice memo."

Testing Your Skill

If an agent doesn't invoke your skill, it's almost always because the description didn't match how the user phrased their request.

Test against multiple phrasings:

• "transcribe this audio" ✓
• "convert this to text" ✓
• "what does this recording say" ✓
• "STT on this file" ✓

Skill Body

After the frontmatter, include:

• When to Use section with bullet points
• Available Commands — list the commands the skill exposes through the runner
• Instructions for how to accomplish the task
• Examples showing anima skill run … invocations
• Notes for edge cases or important details

Writing Scripts that Work with the Runner

1. Take absolute paths in argv. Don't infer paths from CWD. The runner sets

cwd to SKILL_DIR for safety, but explicit absolute paths are clearer.

1. Use process.env.SKILL_DIR for any skill-internal resource (data files,

playlists, prompt templates) instead of relative paths or process.cwd().

1. Report progress for long tasks via anima scheduler update_progress

when ANIMATASKID is set (no-op when not):

``js const { spawnSync } = require("childprocess"); function progress(message) { if (!process.env.ANIMATASKID) return; spawnSync( "anima", [ "scheduler", "updateprogress", "--taskId", process.env.ANIMATASKID, "--message", message, ], { stdio: "ignore", timeout: 5000 }, ); } ``

1. Validate required env vars early and exit with a clear error. The runner

also checks commands.*.env from skill.json before exec.

Template

---
name: doing-something
description: "MUST be used when [trigger condition]. [What it does]. Triggers on: [phrase1], [phrase2], [phrase3], [phrase4], [phrase5]."
---

# Doing Something

Use this skill when the user wants to [goal].

## When to Use

- [Scenario 1]
- [Scenario 2]
- [Scenario 3]

## Available Commands

This skill is invoked through the **anima skill runner**.

- **`<command-name>`** — description of what the command does

Inspect:

\`\`\`bash
anima skill help doing-something <command-name>
\`\`\`

## Instructions

1. [Step 1]
2. [Step 2]
3. [Step 3] — `anima skill run doing-something <command> <absolute-path-arg>`

## Examples

\`\`\`bash
anima skill run doing-something <command> /absolute/path/to/input
\`\`\`

## Notes

- [Important note 1]
- [Important note 2]

skill.json template

{
  "id": "doing-something",
  "description": "Short description shown in `anima skill list`.",
  "commands": {
    "<command-name>": {
      // Choose ONE: `script` for skill-local logic, `command` for a PATH binary
      "script": "scripts/<command-name>.<ext>",
      // OR
      "command": "<binary-name>",

      "runtime": "node | bun | python3 | bash",
      "longRunning": false,
      "description": "What the command does.",
      "args": [
        {
          "name": "arg-name",
          "type": "absolute-file | absolute-folder | string | number | boolean",
          "required": true,
          "description": "What this argument is for"
        }
      ],
      "env": ["REQUIRED_ENV_VAR_1", "REQUIRED_ENV_VAR_2"],
      "timeoutMs": 600000
    }
  }
}