skill-engineer

# Skill Engineer

Own the full lifecycle of agent skills in your OpenClaw agent kit. The entire multi-agent workflow depends on skill quality — a weak skill produces weak results across every run.

**Core principle:** Builders don't evaluate their own work. This skill enforces separation of concerns through a multi-agent architecture where design, review, and testing are performed by independent subagents.


## Skill Taxonomy

> Source: Anthropic "Improving skill-creator" (2026-03-03)

Skills fall into two categories. This distinction drives design decisions, testing strategy, and lifecycle management.

### Capability Uplift Skills

The model can't do it well alone — the skill injects techniques, patterns, or constraints that produce better output than prompting alone.

**Examples:** Document creation skills (PDF generation), complex formatting, specialized analysis pipelines.

**Testing focus:** Monitor whether the base model has caught up. If the base model passes your evals *without* the skill loaded, the skill's techniques have been incorporated into model default behavior. The skill isn't broken — it's no longer necessary.

**Lifecycle:** These skills may "retire" as models improve. Build evals that can detect when retirement is appropriate.

### Encoded Preference Skills

The model can already do each step — the skill sequences operations according to your team's specific process.

**Examples:** NDA review against set criteria, weekly report generation from specific data sources, brand compliance checks.

**Testing focus:** Verify the skill faithfully reproduces your actual workflow, not the model's "free improvisation." Fidelity to process is the metric.

**Lifecycle:** These skills are durable — they encode organizational knowledge that doesn't change with model capability. They need maintenance when *processes* change, not when *models* change.

### Design Implication

When the Designer begins work, classify the skill:

| Classification | Design priority |...

# skill-engineer

Design, test, review, and maintain agent skills for OpenClaw systems. Provides comprehensive lifecycle management for agent skills, including validation, quality assurance, integration testing, and — new in v3.1.1 — README sync to ensure documentation reflects the actual shipped skill.

## Overview

This skill enables systematic development and maintenance of agent skills following Anthropic's best practices. It includes:

- **Progressive disclosure** design patterns (YAML frontmatter → SKILL.md → references/)
- **Quality scoring system** for skill evaluation (completeness, clarity, balance, integration)
- **Tool selection validation** to prevent common execution errors
- **Self-play testing** protocols for skill validation
- **Agent kit audit** procedures for maintaining skill ecosystem health
- **README sync step** — after quality gates pass, README is regenerated from the final implementation to ensure accuracy before push

## Requirements

This skill uses a **multi-agent architecture** with three roles: Designer, Reviewer, and Tester. The orchestrating agent must be able to **spawn subagents** (e.g., via `sessions_spawn` in OpenClaw).

**Minimum setup:**
- An OpenClaw agent with subagent spawning capability (main session or top-level agent)
- At least 3 subagent sessions available per skill design cycle

**Single-agent fallback:** If subagent spawning is unavailable, the skill can run in role-based mode where one agent switches between Designer/Reviewer/Tester phases sequentially.

## Workflow

```
Requirements gathering
    ↓
Designer → Reviewer ──pass──→ Tester ──pass──→ README Sync → Push to GitHub
               │                   │
              fail                fail
               └──── Designer revises (max 3 iterations) ────┘
```

### Step Summary

| # | Step | Who | When to proceed |
|---|------|-----|----------------|
| 1-4 | Gather requirements, spawn Designer | Orchestrator | Always |
| 5-6 | Review artifacts | Reviewer | After...