| name | skill-engineer |
| slug | skill-engineer |
| version | 3.1.0 |
| description | Design, test, review, and maintain agent skills for OpenClaw systems using multi-agent iterative refinement. Orchestrates Designer, Reviewer, and Tester subagents for quality-gated skill development.
Use when: user asks to "design skill", "review skill", "test skill", "audit skills", "refactor skill", or mentions "agent kit quality".
|
| triggers | ["design skill","review skill","test skill","audit skills","refactor skill","agent kit quality","build new skill","create skill"] |
| metadata | {"openclaw":{"emoji":"🛠️","requires":{"bins":["git"]}}} |
Skill Engineer
Design, test, review, and maintain agent skills using multi-agent iterative refinement.
Quick Start
User: "Create a skill for scraping websites"
1. Gather requirements (problem, audience, inputs/outputs)
2. Spawn Designer → creates SKILL.md
3. Spawn Reviewer → evaluates quality (score ≥28/33)
4. Spawn Tester → validates functionality
5. If pass → deliver to user
6. If fail → iterate (max 3 times)
Architecture
Orchestrator (main agent)
│
├─→ Designer (creates artifacts)
├─→ Reviewer (scores quality)
└─→ Tester (validates functionality)
Success Criteria
- Quality gates - Reviewer scores ≥28/33 (Deploy threshold)
- No blocking issues - Tester reports no "blocking" issues
- All artifacts - SKILL.md, README.md, tests/, scripts/
- OPSEC clean - No hardcoded secrets or private URLs
- Trigger accuracy - ≥90% true positives
Workflow
1. Gather Requirements
| Input | Description |
|---|
| Problem | What capability to enable |
| Audience | Which agent(s) will use |
| Interactions | APIs, files, MCP servers |
| Inputs/Outputs | Data flow |
| Constraints | Performance, security |
2. Design Phase
- Spawn Designer with requirements
- Get SKILL.md, skill.yml, README.md
- Present 3-5 name candidates to user
3. Review Phase
- Spawn Reviewer with artifacts
- Quality rubric: 33 checks across 8 categories
- Scoring: 28-33 = Deploy, 20-27 = Revise, <20 = Fail
4. Test Phase
- Spawn Tester with artifacts
- Run self-play validation
- Check trigger accuracy
5. Iterate (max 3 times)
- If fail → Designer revises
- Track iteration count
- After 3 failures → report to user
Quality Rubric (33 Points)
| Category | Max | Description |
|---|
| SQ-A Completeness | 7 | All required sections present |
| SQ-B Clarity | 5 | Unambiguous, well-written |
| SQ-C Balance | 4 | AI/Script split appropriate |
| SQ-D Integration | 4 | Works with standard kit |
| SCOPE | 3 | Clean boundaries |
| OPSEC | 2 | No security violations |
| REF | 3 | Sources cited |
| ARCH | 2 | Separation of concerns |
| TOTAL | 30 | |
Subagent Roles
Designer
- Creates skill artifacts
- Follows progressive disclosure
- Applies scoping rules
Reviewer
- Independent quality evaluation
- Scores against rubric
- Identifies issues + strengths
Tester
- Self-play validation
- Trigger testing
- Functional validation
Error Handling
| Failure | Response |
|---|
| Git push fails | Retry once, then report |
| OPSEC violation | Fail immediately |
| 3 iterations fail | Report to user, stop |
| File write error | Report, fail workflow |
Example
User: "I need a skill for analyzing competitor websites"
Orchestrator:
1. Gather requirements → Problem, audience, interactions
2. Spawn Designer → Creates skill-artifacts
3. Spawn Reviewer → Scores 26/33, "Revise"
4. Feed back to Designer → Fix issues
5. Review again → Score 30/33, "Pass"
6. Spawn Tester → No blocking issues
7. Deliver artifacts
Related Skills
- skill-creator - Simple skill creation
- openclaw-agent-optimize - Agent optimization
- compound-engineering - Self-improvement
References
references/designer-guide.mdreferences/reviewer-rubric.mdreferences/tester-protocol.md
Version: 3.1.0
Based on: Anthropic Complete Guide to Building Skills for Claude (2026-01)
