skill-reviewer-pro（技能审核员专业版）

Skill Reviewer Pro

Comprehensive skill review and validation for OpenClaw skills with scoring rubric, defect checklists, functional verification, optimization recommendations, and workflow validation.

Language Adaptation

IMPORTANT: Always respond in the same language as the user's request.

• If user asks in Chinese → respond in Chinese
• If user asks in English → respond in English
• If user asks in other languages → respond in that language

Auto-adapt to the user's language to ensure clear communication and better user experience.

Review: agent-builder-plus (Example)

Level 1: Format Validation ✅

• package_skill.py validation passed
• YAML frontmatter valid
• File structure correct

Level 2: Writing Quality Assessment

Structure: ✅ Good (6/6)

• Clear frontmatter with name and description
• Well-organized body with workflow sections
• Progressive disclosure pattern used

Description: ✅ Strong (8/8)

• Starts with what skill does: "Build high-performing OpenClaw agents end-to-end."
• Includes trigger: "Use when you want to design a new agent..."
• Specific scope: mentions SOUL.md, IDENTITY.md, AGENTS.md, USER.md, HEARTBEAT.md
• Reasonable length
• Contains searchable keywords

Example Density: ⚠️ Low (1/3)

• SKILL.md has few code blocks
• Could benefit from more concrete examples

Organization: ✅ Good (6/6)

• Organized by workflow (Phase 1 → Phase 2 → Phase 3 → Phase 4)
• Logical flow from interview to generation to validation
• Sections are self-contained

Actionability: ✅ Good (10/10)

• Instructions are imperative
• Steps are ordered logically
• Guardrails checklist provided
• Acceptance tests included

Level 3: Functional Verification ❌ CRITICAL (0/20)

Problem: Generated AGENTS.md template does not match OpenClaw specifications

Missing from templates.md AGENTS.md:

• Session startup requirements (Read SOUL.md, USER.md, memory files)
• Memory workflow details (daily notes vs long-term)
• Group chat behavior guidelines (when to speak, when to stay silent)
• Heartbeat mechanism details (heartbeat vs cron, what to check)
• Tool usage guidelines (Skills provide tools, Voice storytelling, Platform formatting)

Impact: Skills created with agent-builder-plus will be missing critical OpenClaw agent behaviors.

Level 4: Best Practices Check ✅ Good (15/15)

Progressive Disclosure: ✅ Good (5/5)

• SKILL.md is concise
• Detailed information in references/
• References properly linked

Resource Organization: ✅ Good (5/5)

• Only necessary directories created
• No extraneous files

Triggering Accuracy: ✅ Good (5/5)

• Description clearly states when to use
• No "When to Use This Skill" in body

Level 5: Optimization Recommendations

🔴 Critical (Must Fix)

1. Fix AGENTS.md Template Functional Mismatch Update references/templates.md to include all required sections from actual OpenClaw AGENTS.md.

🟡 Major (Should Fix)

2. Add More Concrete Examples Add 3-5 concrete code examples in workflow sections.

3. Add Quick Start Section Add a "Quick Start" section at the beginning of SKILL.md.

Level 6: Workflow Validation ❌ CRITICAL (0/20)

Note: This level applies to workflow tools (skills that guide users through multi-step processes).

Workflow Completeness Check

WORKFLOW COMPLETENESS CHECK:

[ ] Each phase has clear operational steps
[ ] Correct CLI commands are used
[ ] Error handling is documented
[ ] Backup and recovery mechanisms exist
[ ] Verification steps are included
[ ] Configuration management is documented
[ ] Failure recovery strategies exist

agent-builder-plus Analysis:

Detailed Issues

Issue 1: Missing BOOTSTRAP.md in file list

• Phase: Phase 2
• Problem: BOOTSTRAP.md is mentioned in AGENTS.md specs but not in file list
• Impact: New agents lack first-run guidance
• Fix: Add BOOTSTRAP.md to Phase 2 file list

Issue 2: No agent registration logic

• Phase: Missing
• Problem: No openclaw agents add command documented
• Impact: Users must manually configure openclaw.json
• Fix: Add Phase 2.5 with openclaw agents add command

Issue 3: No auth-profiles.json documentation

• Phase: Missing
• Problem: No authentication configuration documentation
• Impact: Agents cannot access external services (e.g., Feishu)
• Fix: Add authentication section after Phase 2.5

Issue 4: No backup mechanism

• Phase: Missing
• Problem: No backup before modifying openclaw.json
• Impact: Config errors can break entire OpenClaw system
• Fix: Add backup step before Phase 2.5

Issue 5: No configuration verification

• Phase: Missing
• Problem: No verification after agent registration
• Impact: Agent may not start correctly
• Fix: Add Phase 2.6 with verification steps

Issue 6: No directory creation robustness

• Phase: Phase 2
• Problem: No error handling for directory creation
• Impact: File writes fail silently if directory creation fails
• Fix: Add mkdir -p with error checking

Issue 7: No automated testing (Phase 8)

• Phase: Missing
• Problem: Only Phase 4 (acceptance tests), no automated testing
• Impact: No complete testing workflow
• Fix: Add Phase 8 with automated test commands

Issue 8: No failure recovery strategies

• Phase: All phases
• Problem: No recovery strategies for any phase
• Impact: Users don't know how to recover from failures
• Fix: Add failure recovery section to each phase

Issue 9: Unclear Feishu binding method

• Phase: Phase 1
• Problem: Question 2 asks "Which channels?" but doesn't clarify Feishu binding
• Impact: Users unclear about channel binding vs /agentname command
• Fix: Clarify Feishu binding options in Phase 1

Issue 10: Optional systemd heartbeat

• Phase: Optional
• Problem: No systemd service configuration option
• Impact: Cannot run agent as systemd service
• Fix: Add optional systemd service configuration

Score Summary

SKILL REVIEW SCORECARD
═══════════════════════════════════════
Skill: agent-builder-plus
Reviewer: 大鱼
Date: 2026-02-27

Category              Score    Max
─────────────────────────────────────
Level 1: Format       10       10  ✅
Level 2: Writing
  - Structure          6        6  ✅
  - Description        8        8  ✅
  - Example density    1        3  ⚠️
  - Organization       6        6  ✅
  - Actionability     10       10  ✅
Level 3: Functional    0       20  ❌ CRITICAL
Level 4: Best Practices
  - Progressive       5        5  ✅
  - Resources         5        5  ✅
  - Triggering        5        5  ✅
Level 5: Optimization  0       10  (recommendations provided)
)
Level 6: Workflow      0       20  ❌ CRITICAL
─────────────────────────────────────
TOTAL                 56       116

RATING: 56/116 = Fair — significant gaps to address
VERDICT: REWORK (functional and workflow verification failed)

Top Defects

1. Functional Mismatch: AGENTS.md template missing critical OpenClaw specifications.

2. Workflow Incomplete: Missing agent registration, configuration verification, backup mechanisms, error handling, and failure recovery strategies.

Recommendation

1. Update references/templates.md AGENTS.md template with all required sections
2. Add Phase 2.5 (agent registration with openclaw agents add)
3. Add Phase 2.6 (configuration verification)
4. Add backup mechanism before config changes
5. Add failure recovery strategies to each phase
6. Add directory creation error handling
7. Add Phase 8 (automated testing)
8. Clarify Feishu binding options in Phase 1
9. Add optional systemd service configuration

Review Workflow Guide

Level 1: Format Validation

Run automatic validation using package_skill.py:

python3 /home/yupeng/.npm-global/lib/node_modules/openclaw/skills/skill-creator/scripts/package_skill.py <skill-path>

Level 2: Writing Quality Assessment

Check structure, description, examples, organization, and actionability.

Level 3: Functional Verification

For OpenClaw skills, compare generated templates with actual specifications.

Level 4: Best Practices Check

Verify progressive disclosure, resource organization, and triggering accuracy.

Level 5: Optimization Recommendations

Provide prioritized recommendations (Critical/Major/Minor).

Level 6: Workflow Validation (For Workflow Tools)

Workflow Completeness Check

WORKFLOW COMPLETENESS CHECK:

[ ] Each phase has clear operational steps
[ ] Correct CLI commands are used
[ ] Error handling is documented
[ ] Backup and recovery mechanisms exist
[ ] Verification steps are included
[ ] Configuration management is documented
[ ] Failure recovery strategies exist

Phase-by-Phase Validation

For each phase in the workflow:

PHASE VALIDATION CHECKLIST:

Phase: [phase name]

Operational Steps:
[ ] Clear, step-by-step instructions provided
[ ] Commands are accurate and complete
[ ] Required parameters are specified

Error Handling:
[ ] Common errors are documented
[ ] Recovery steps are provided
[ ] Error messages are explained

Verification:
[ ] Success criteria are defined
[ ] Verification steps are included
[ ] How to confirm completion is explained

Dependencies:
[ ] Prerequisites are listed
[ ] Order dependencies are clear
[ ] Previous phase completion is verified

CLI Command Validation

For workflow tools that use OpenClaw CLI commands:

CLI COMMAND VALIDATION:

[ ] Commands are verified with `openclaw <command> --help`
[ ] Command syntax is correct
[ ] Required options are documented
[ ] Default values are explained
[ ] Example commands are runnable

Configuration Management

For tools that modify OpenClaw configuration:

CONFIGURATION MANAGEMENT CHECK:

[ ] Files being modified are listed
[ ] Backup mechanism is provided
[ ] Rollback steps are documented
[ ] Configuration validation is included
[ ] Authentication requirements are explained

Common Defects

Critical

• Invalid frontmatter
• Broken code examples
• Misleading description
• Functional mismatch (OpenClaw skills)
• Workflow incomplete (missing phases, no CLI commands)
• No backup before config changes

Major

• No "When to Use" information
• Text walls without examples
• Examples missing language tags
• Abstract organization
• Duplication between SKILL.md and references
• No error handling
• No verification steps
• No failure recovery strategies

Minor

• Placeholder values
• Inconsistent formatting
• Extraneous files
• Missing cross-references
• No quick start guide

Language Adaptation Guidelines

CRITICAL: Always respond in the same language as the user's request.

Detection Rules

1. Detect user language from the request message

- Check the language of the user's input message - Use the same language for all responses

1. Language mapping

- Chinese (中文/汉语) → Respond in Chinese - English → Respond in English - Japanese (日本語) → Respond in Japanese - Korean (한국어) → Respond in Korean - Other languages → Respond in the detected language

1. Consistency

- Once language is detected, use it for the entire review - All section headers, descriptions, and feedback should be in the same language - Technical terms (like "Level 1", "Level 2") can remain in English if they are standard terminology

Example Scenarios

Scenario 1: User asks in Chinese

User: 请审查一下 agent-builder-plus 这个 skill
AI: (responds in Chinese)
## 审查结果：agent-builder-plus
...

Scenario 2: User asks in English

User: Please review the agent-builder-plus skill
AI: (responds in English)
## Review Results: agent-builder-plus
...

Scenario 3: User asks in mixed language

User: 请审查 agent-builder-plus skill
AI: (responds in Chinese, as the primary language is Chinese)
## 审查结果：agent-builder-plus
...

Implementation Notes

• Language detection should be done at the start of the review
• Use simple language detection (check for Chinese characters, etc.)
• If language cannot be detected, default to the language of the previous interaction
• Maintain language consistency throughout the entire review process

Technical Terms

Keep technical terms in English when appropriate:

• Skill names (e.g., "agent-builder-plus")
• File names (e.g., "SKILL.md", "IDENTITY.md")
• CLI commands (e.g., "openclaw agents add")
• Technical concepts (e.g., "workspace", "agent", "channel")

But translate descriptions, feedback, and explanations to the user's language.