Prompt Manager

Purpose

Optimize AILANG teaching prompts for maximum conciseness and accuracy while maintaining AI code generation effectiveness.

Key Changes (October 2025)

Before

• Skill focused on version management and workflow
• No emphasis on prompt optimization
• No automated analysis tools
• Verbose documentation

After

• Primary mission: Optimize prompts for token efficiency
• Target: <4000 words per prompt (currently 4358 words)
• Automated analysis: identify optimization opportunities
• Progressive disclosure: Detailed guides in resources/

What This Skill Does Now

1. Analyzes Prompts for Bloat

.claude/skills/prompt-manager/scripts/analyze_prompt_size.sh prompts/v0.3.16.md

Identifies:

• Word count vs target (<4000)
• Code block proliferation (60 blocks → target: 5-10)
• Missing tables (0 → target: 10+)
• Section-by-section word counts
• Optimization opportunities

2. Guides Optimization Strategy

Key techniques taught:

• Replace prose with tables (-67% tokens)
• Consolidate scattered examples (-56% tokens)
• Link to external docs (-76% tokens)
• Create quick reference section
• Remove redundancy

3. Validates Optimizations

Ensures:

• Token reduction: 30-50%
• Eval success rate maintained
• All external links resolve
• Examples still work

File Structure

.claude/skills/prompt-manager/
├── skill.md                              # Main skill (concise, optimization-focused)
├── scripts/
│   ├── create_prompt_version.sh          # Version management
│   ├── update_hash.sh                    # Hash integrity
│   └── analyze_prompt_size.sh            # NEW: Optimization analysis
└── resources/
    ├── prompt_optimization.md            # Detailed optimization strategies
    └── workflow_guide.md                 # Detailed workflow examples

Progressive Disclosure

1. skill.md (always loaded) - Concise workflow + optimization principles
2. prompt_optimization.md (load when optimizing) - Detailed strategies, before/after examples
3. workflow_guide.md (load when needed) - Detailed workflows, troubleshooting
4. Scripts (execute as needed) - Automation tools

Current Prompt Analysis (v0.3.16)

Total words: 4358 (target: <4000) - 8% over
Total lines: 1214 (target: <200) - 507% over
Code blocks: 60 (target: 5-10) - 500% over
Tables: 0 (target: 10+) - missing

Optimization potential: ~50% token reduction

Optimization Roadmap

Phase 1: Quick Wins (v0.3.17)

• [ ] Convert builtin docs to tables (~1000 tokens saved)
• [ ] Consolidate 60 code blocks to 8-10 comprehensive examples (~500 tokens)
• [ ] Add quick reference section at top
• [ ] Target: 3500 words (-20%)

Phase 2: Deep Optimization (v0.3.18)

• [ ] Move type system details to docs/guides/types.md (~800 tokens)
• [ ] Move module system details to docs/guides/modules.md (~500 tokens)
• [ ] Create effect system guide, link from prompt (~400 tokens)
• [ ] Target: 2500 words (-43%)

Phase 3: Maximum Density (v0.4.0)

• [ ] Reference ailang builtins list instead of duplicating
• [ ] Single comprehensive example covering all features
• [ ] Quick reference + external links structure
• [ ] Target: <2000 words (-54%)

Success Metrics

Prompt profile v0.3.16 → v0.4.0:

• Words: 4358 → <2000 (-54%)
• Lines: 1214 → <200 (-84%)
• Code blocks: 60 → 8-10 (-83%)
• Tables: 0 → 10+ (new)
• External links: minimal → 10+ (new)

Validation:

• Eval success rate: maintained or improved
• AI response time: improved (less token processing)
• Developer experience: faster to read and understand

Integration

Works with:

• eval-analyzer: Verify prompt accuracy before optimization
• post-release: Run baselines after optimization to validate
• ailang builtins: Reference instead of duplicating builtin docs
• docs/guides/: External docs for detailed explanations

Usage Example

# Analyze current prompt
.claude/skills/prompt-manager/scripts/analyze_prompt_size.sh prompts/v0.3.16.md

# Create optimized version
.claude/skills/prompt-manager/scripts/create_prompt_version.sh v0.3.17 v0.3.16 "Optimize for conciseness (-30% tokens)"

# Apply optimization strategies (refer to resources/prompt_optimization.md)
# - Convert builtin docs to tables
# - Consolidate 60 examples to 8 comprehensive ones
# - Link type/module system details to docs/

# Validate
.claude/skills/prompt-manager/scripts/analyze_prompt_size.sh prompts/v0.3.17.md
# Should show: <3500 words, 10+ tables, 8-10 code blocks

# Verify accuracy
.claude/skills/eval-analyzer/scripts/verify_prompt_accuracy.sh v0.3.17

# Update hash
.claude/skills/prompt-manager/scripts/update_hash.sh v0.3.17

# Test effectiveness
ailang eval-suite --models gpt5-mini --output eval_results/test_v0.3.17
ailang eval-compare eval_results/baselines/v0.3.16 eval_results/test_v0.3.17

# Commit if success rate maintained
git add prompts/v0.3.17.md prompts/versions.json
git commit -m "feat: Optimize v0.3.17 prompt for conciseness (-30% tokens)"

Anti-Patterns (What This Skill Prevents)

❌ Verbose builtin documentation (use tables + reference ailang builtins list) ❌ 60 scattered examples (consolidate to 8-10 comprehensive) ❌ Explaining "why" in prompt (move to design docs) ❌ Historical context (move to CHANGELOG.md) ❌ Implementation details (link to code) ❌ Apologetic limitations (be direct)

References

• Optimization strategies: resources/prompt_optimization.md
• Detailed workflows: resources/workflow_guide.md
• Main skill: skill.md