Get the FREE Ultimate OpenClaw Setup Guide →

touch-skill

npx machina-cli add skill Vibe-Builders/claude-prime/touch-skill --openclaw
Files (1)
SKILL.md
5.3 KB

Ultrathink.

Role

You are a skill architect. Create or update skills — not features or existing code.

Context Assessment

  • Similar skill exists? → Suggest enhancing instead
  • Requirements already gathered? → Skip research

Core Principles

  1. Skills are not documentation — Active workflow knowledge, not passive reference
  2. Soft line limits — SKILL.md ~150, references ~500, workflows ~200
  3. Progressive disclosure — Load details on-demand via references/workflows
  4. Script-first — Executable scripts for deterministic, token-efficient operations
  5. SKILL.md = philosophy, not patterns — Principles (WHY) inline, details (HOW) in references
  6. Link, don't embed — Cross-reference with "-> See [file.md]" pattern
  7. Right-altitude — Concrete but not hardcoded
  8. Anti-convergence — List what to avoid, suggest uncommon alternatives

Process

1. Understand

  • Parse requirements from $ARGUMENTS
  • Identify skill purpose, scope, and type (knowledge vs workflow)

2. Research

Three sources make a good skill. Follow skill-research.md.

  1. Discover existing skills — Search skill-sources.md. Skim SKILL.md first to check scope relevance. If relevant, read references and produce an extraction note (reuse/skip/gaps). If not relevant, state why and move on.
  2. Fetch official docs — Combine with research and docs-seeker skill. Extract correct API patterns, vendor recommendations, and migration guides.
  3. Synthesize — Combine existing skill patterns with official docs into right-altitude domain knowledge. Skills teach how to use a technology well, not project conventions. Flag any content from model knowledge (no doc backing).

If user provides URLs → explore internal links for additional content.

3. Design

GATE: User approves skill design before creation.

4. Create

Create at .claude/skills/{skill-name}/:

  • SKILL.md — main entry point
  • references/ — supporting documentation
  • scripts/ — executable utilities (if needed)
  • templates/ — reusable templates (if needed)

Quick start scripts:

scripts/init_skill.py <skill-name> --path <output-directory>
scripts/init_workflow_skill.py <skill-name> --path <output-directory>

5. Validate

scripts/validate_skill.py <path/to/skill>
scripts/validate_skill.py <path/to/skill> --type=workflow

Skill Types

Knowledge/Reference Skills

Workflow Skills

Task runners (/cook, /fix, /discuss) with roles, processes, gates, and $ARGUMENTS.

References

ReferenceContent
anatomy.mdSkill structure, YAML frontmatter
requirements.mdLine limits, naming, .env rules
bundled-resources.mdscripts/, references/, assets/
design-principles.mdPhilosophy, progressive disclosure
workflow-patterns.mdGates, delegation, role boundaries
examples.mdGood/bad patterns
skill-sources.mdCurated external skill sources and repositories

Scripts

ScriptPurpose
init_skill.pyInitialize new skill from template
init_workflow_skill.pyInitialize workflow-type skill
validate_skill.pyValidate skill structure (--type=workflow for workflow checks)
package_skill.pyCreate distributable zip

Templates

TemplatePurpose
skill-template.mdKnowledge/reference SKILL.md
workflow-skill-template.mdWorkflow/task-runner SKILL.md
script-template.pyExample Python script
reference-template.mdExample reference file

Workflows

Requirements

<user-prompt>$ARGUMENTS</user-prompt>

Source

git clone https://github.com/Vibe-Builders/claude-prime/blob/main/.claude/skills/touch-skill/SKILL.mdView on GitHub

Overview

Touch-skill provides a structured approach to designing and updating agent skills. It emphasizes practical templates, rigorous validation, and progressive disclosure so your SKILL.md remains concise while supporting deep references.

How This Skill Works

The process follows Understand, Research, Design, Create, and Validate, leveraging templates and references. It promotes a script-first, reference-led workflow to ensure deterministic, token-efficient skill creation and robust validation before deployment.

When to Use It

  • When creating a new skill from scratch
  • When refactoring an existing skill to align with templates and standards
  • When validating a skill against requirements before deployment
  • When composing SKILL.md with progressive disclosure via references
  • When designing knowledge or workflow skills using the appropriate templates

Quick Start

  1. Step 1: Understand requirements from $ARGUMENTS and identify if the skill is knowledge or workflow
  2. Step 2: Design SKILL.md using the appropriate template and plan references
  3. Step 3: Create the skill under .claude/skills/{skill-name}/ and run validation scripts

Best Practices

  • Use the correct template (skill-template.md for knowledge, workflow-skill-template.md for workflows)
  • Keep SKILL.md concise (target ~150 words for the main file) and push deep details to references
  • Link to references and sources rather than embedding full content in SKILL.md
  • Include executable tooling in scripts/ to support deterministic validation and setup
  • Obtain GATE approval from the user before creating the skill design

Example Use Cases

  • Create a new data-cleanse skill from scratch following touch-skill guidance and store in .claude/skills/data-cleanse/
  • Refactor an existing fetch-report skill to align with the templates and design principles
  • Validate a candidate summarize skill against requirements.md using validate_skill.py
  • Develop a workflow skill using workflow-skill-template.md and workflow-patterns.md for patterns
  • Add progressive disclosure by linking to references/anatomy.md and references/design-principles.md

Frequently Asked Questions

Add this skill to your agents
Sponsor this space

Reach thousands of developers