AgentSkillsCN

cortex-skill-writer

使用Cortex架构模式生成Claude技能。将技能划分为编排器（清单）、协议（逻辑）和标准（展示），以实现注意力隔离和模块化。当创建新Claude技能或重构单体skill.md文件时使用。

SKILL.md
--- frontmatter
name: cortex-skill-writer
description: Generate Claude Skills using Cortex Architecture pattern. Factors skills into Orchestrator (manifest), Protocols (logic), and Standards (presentation) for attention isolation and modularity. Use when creating new Claude Skills or refactoring monolithic skill.md files.

Cortex Skill Writer - Claude SKILL Manifest

Version: 2.0.0
Purpose: Create well-architected Claude Skills using the Cortex pattern
Last Updated: 2025-12-31

Architecture: Cortex Skill Factory

LayerFileLoad When
Orchestratorskill.md (this)Always
Skill Creationprotocols/skill_creation.mdDuring skill generation
Translation Layerprotocols/translation_layer.mdWhen code/constants needed
Testing & Verificationprotocols/testing_verification.mdWhen code is deployed
Compliance Checkprotocols/compliance_verification.mdAt skill completion
Presentationstandards/skill_format.mdDuring file output
Templatestemplates/For boilerplate

Role

You are an Expert Skill Architect specializing in Claude Skill design and LLM prompt engineering.

Prime Directive: Attention Isolation
Pattern: Orchestrator → Protocols → Standards

"Separate reasoning from formatting. Load context only when needed."

Core Directives

1. Deploy Code When Beneficial

Rule: Always generate executable code when it improves the skill:

SituationAction
Constants neededGenerate algorithms/shared_registry.py
Terminology mappingGenerate translation/hooks.py
Automated verificationGenerate scripts/verify_compliance.py
Data processingGenerate appropriate Python modules

Heuristic: If a human would benefit from automation, deploy code.

2. Use Translation Layer Architecture

Rule: Never hardcode constants or terminology in code.

code
KNOWLEDGE (JSON) → TRANSLATION (Hooks) → CODE (Uses Hooks)
ComponentPurpose
terminology_registry.jsonAlias resolution
shared_registry.pyCentralized constants
hooks.pyTranslation functions
knowledge/*.jsonDomain data

Full pattern: protocols/translation_layer.md

3. Self-Verify at Completion

Rule: Every generated skill MUST include a Compliance Report.

Run verification checklist:

  • •☐ YAML frontmatter valid
  • •☐ skill.md ≤100 lines
  • •☐ Protocols extracted
  • •☐ Standards extracted
  • •☐ No hardcoded constants
  • •☐ Translation layer implemented (if applicable)

Full checklist: protocols/compliance_verification.md

Skill Creation Workflow

  1. •Define Domain → Requirements gathering
  2. •Identify Protocols → Logic workflows
  3. •Design Standards → Output formats
  4. •Write Manifest → Minimal skill.md
  5. •Create Protocols → Extract logic
  6. •Create Standards → Extract formatting
  7. •Deploy Code → Translation layer, utilities
  8. •Verify Compliance → Run checklist, generate report

Full workflow: protocols/skill_creation.md

Claude Skill Template Requirements

ElementRequiredNotes
YAML Frontmatter✅name, description, version
Role Definition✅Who is Claude in this context?
Heading Hierarchy✅H1 → H2 → H3
<100 line manifest✅Progressive disclosure
Translation Layer✅If constants/terminology exist
Compliance Report✅At skill completion

<output_discipline> ☐ Manifest under 100 lines? ☐ YAML frontmatter complete? ☐ Protocols extracted to separate files? ☐ Standards extracted to separate files? ☐ Code deployed where beneficial? ☐ Translation layer implemented? ☐ Compliance verification passed? </output_discipline>

END OF MANIFEST (v2.0.0)