Staged Content Refinement

Process extracted skills and docs from multiple projects into consolidated, project-agnostic patterns.

Installation

OpenClaw / Moltbot / Clawbot

npx clawhub@latest install refinement

Before Starting

1. Verify staging has content from multiple projects:

• ai/staging/skills/ — Extracted skills
• ai/staging/docs/ — Extracted methodology docs

1. Load quality criteria: ../extraction/references/skill-quality-criteria.md

Refinement Process

Phase 1: Inventory

Scan all staged content and catalog what exists.

For skills:

ai/staging/skills/
├── [project-a]-[category]/
├── [project-b]-[category]/
└── ...

For docs:

ai/staging/docs/
├── [project-a]-summary.md
├── [project-a]-design-system.md
├── [project-b]-summary.md
└── ...

Create inventory:

Phase 2: Pattern Analysis

Identify patterns that appear across multiple projects.

Look for:

1. Design system patterns (highest priority)

• Common token structures
• Similar aesthetic approaches
• Shared color/typography patterns

1. Architecture patterns

• Similar folder structures
• Common component organization
• Shared data flow patterns

1. Workflow patterns

• Common Makefile targets
• Similar CI/CD approaches
• Shared dev setup patterns

1. Methodology insights

• Decision-making patterns
• Trade-offs that appear repeatedly
• Common anti-patterns discovered

Questions to answer:

• What patterns appear in 2+ projects?
• What's project-specific vs generalizable?
• What methodology insights can update the main docs?

Phase 2.5: Conflict Detection

Before consolidating, check for conflicts with existing skills.

Scan existing skills:

ls ai/skills/*/

For each staged skill, check:

Conflict resolution:

Run overlap check:

/check-overlaps

Document conflicts:

• Note which existing skills overlap
• Decide: merge into existing vs create new
• Update existing skills if merging

Phase 3: Consolidation

Merge similar patterns into unified, project-agnostic skills.

For each pattern group:

1. Identify the core — What's common across all instances?
2. Strip project specifics — Remove project names, specific values
3. Generalize — Make it applicable to any project
4. Enhance — Add insights learned from seeing multiple implementations

Example consolidation:

Staged:
├── project-a-design-system/  (retro-futuristic tokens)
├── project-b-design-system/  (video game UI tokens)
└── project-c-design-system/  (minimal dark tokens)

Consolidated:
└── design-system-patterns/   (how to build distinctive design systems)
    - Token structure patterns
    - Aesthetic direction documentation
    - Common anti-patterns across projects

Consolidation rules:

Phase 4: Methodology Updates

Extract insights to update the main methodology docs.

Review staged docs for:

1. New patterns not in current methodology
2. Refinements to existing approaches
3. Anti-patterns discovered across projects
4. Decision rationale worth capturing

Update locations:

Phase 5: Promotion

Move refined skills to active locations.

Skill promotion:

# From staging to active
mv ai/staging/skills/[refined-skill]/ ai/skills/[refined-skill]/

Doc updates:

• Update main docs with methodology insights
• Archive or delete processed staging content

Post-promotion:

• Verify skills have proper descriptions
• Test skill activation with sample prompts
• Clean up staging folder

Phase 6: Testing

After promotion, verify skills work correctly.

Activation testing: For each promoted skill, test that it activates on expected triggers:

Test: "I need to [trigger from When to Use]"
Expected: Skill should be suggested/activated

Quality verification:

• [ ] Description triggers correctly
• [ ] Code examples are valid and runnable
• [ ] NEVER Do section is actionable
• [ ] Related skills are linked correctly

Rollback if issues:

# If skill has problems, move back to staging
mv ai/skills/[category]/[skill]/ ai/staging/skills/

Document test results:

• Note any activation issues
• Record false positives (activates when shouldn't)
• Record false negatives (doesn't activate when should)

Error Handling

Output Locations

Refined skills go to:

ai/skills/           # Cursor-specific skills
ai/skills/           # Claude Code skills (if applicable)

Methodology updates go to:

docs/
├── METHODOLOGY.md        # Philosophy and approach
├── TECH-STACK.md         # Technology patterns
├── ARCHITECTURE.md       # Skill/agent architecture
└── WORKFLOW.md           # Processes

Staging cleanup:

ai/staging/skills/        # Clear after processing
ai/staging/docs/          # Clear after processing
ai/archive/               # (optional) Keep for reference

Consolidation Examples

Design Systems

Before (3 project-specific skills):

• project-a-design-system — Retro-futuristic tokens
• project-b-design-system — Video game UI tokens
• project-c-design-system — Cyberpunk terminal tokens

After (1 consolidated skill):

• distinctive-design-systems — How to create design systems with personality
• Token structure patterns
• Aesthetic documentation approach
• Anti-patterns (generic bootstrap, "clean" without personality)
• Examples from different aesthetic directions

Architecture

Before (multiple docs):

• project-a-architecture.md
• project-b-architecture.md

After:

• Update docs/ARCHITECTURE.md with common patterns
• Create skill only if patterns are unique enough

NEVER Do

• NEVER keep project-specific details in consolidated skills
• NEVER create redundant skills — Merge similar patterns
• NEVER skip methodology updates — Insights should flow back to docs
• NEVER promote low-quality skills — Must pass quality criteria
• NEVER leave staging cluttered — Clean up after processing

Quality Check

Before finishing refinement:

• [ ] All staged content reviewed?
• [ ] Conflict detection completed (Phase 2.5)?
• [ ] Common patterns identified and consolidated?
• [ ] Consolidated skills are project-agnostic?
• [ ] Methodology docs updated with insights?
• [ ] Refined skills promoted to active locations?
• [ ] Activation testing completed (Phase 6)?
• [ ] Staging folder cleaned up?

Related Skills

• Agent: ai/agents/refinement/ — Autonomous refinement workflow
• Command: /refine-staged — Quick refinement command
• Command: /promote-skill — Single skill promotion
• Previous step: ai/skills/extraction/ — Pattern extraction
• Quality criteria: ../extraction/references/skill-quality-criteria.md