Super Skills
Quick Reference
┌─────────────────────────────────────────────────────────────┐
│ 🎯 Core: Understand → Plan → Execute → Iterate │
│ 🚨 Rule: Problem → Analyze → Solve → Continue (NEVER WAIT) │
│ 📊 Complexity: 4-8 Direct | 9-14 Staged | 15-20 Iterative │
└─────────────────────────────────────────────────────────────┘
Core Philosophy
| Principle | Description |
|---|
| Understand First | Deeply analyze requirements before acting |
| Incremental Delivery | Produce verifiable intermediate results at each step |
| Expect Failures | Design recovery mechanisms for every stage |
| Never Stop | Solve problems proactively; NEVER pause waiting for user |
Workflow
UNDERSTAND ──→ PLAN ──→ EXECUTE ──→ ITERATE
│ │ │ │
Multi-layer Task Search & Feedback
Analysis Breakdown Install Adjustment
Complexity Risk ID Checkpoint Optimize
Phase 1: Understanding
1.1 Multi-Layer Analysis
understanding:
surface: # Literal meaning
request: "{user's exact words}"
keywords: [key terms]
intent: # True intent
goal: "{what user really wants}"
success: "{success criteria}"
context: # Environment
env: "{OS/language/framework}"
constraints: "{time/resources/permissions}"
hidden: # Unstated needs
assumptions: "{implicit assumptions}"
edge_cases: "{boundary conditions}"
1.2 Requirement Handling
| Type | Action |
|---|
| Explicit | Execute directly |
| Implicit | Proactively supplement |
| Ambiguous | State assumptions, choose most likely interpretation |
| Conflicting | Point out conflict, request clarification |
| Out of Scope | Explain why, provide alternatives |
1.3 Complexity Score
Dimensions (1-5 each):
technical + scope + uncertainty + dependencies = total
Strategy:
4-8 → Direct execution
9-14 → Staged verification
15-20 → Iterative prototyping
1.4 Clarification
Confidence ≥70%: State assumptions → Continue → Leave adjustment points
Confidence <70%: Provide options for user to choose (NO open-ended questions)
Phase 2: Planning
2.1 Task Structure
task:
id: 1
name: "{task name}"
capability: "{capability type}"
input: [required inputs]
output: "{type/format}"
depends_on: [prerequisite task IDs]
timeout: "30s"
retries: 3
validation: "{success condition}"
fallback: "{alternative approach}"
2.2 Dependency Graph
[1:Setup] → [2:Auth] → [3:Fetch] ─┐
↓
[4:Process] ← [5:Transform] ← [6:Parse]
↓
[7:Output]
Critical Path: 1→2→3→6→5→4→7
Parallel Opportunities: Independent tasks can run concurrently
2.3 Capability Map
| Capability | Status | Keywords |
|---|
browser_automation
| ❌ | browser, puppeteer, playwright |
api_integration
| ❌ | api, rest, {service} |
data_extraction
| ⚠️ | parse, pdf, ocr |
message_delivery
| ❌ | slack, discord, email |
database_operations
| ❌ | sql, mongodb |
deployment
| ❌ | deploy, docker, k8s |
data_transformation
| ✅ | — |
content_generation
| ✅ | — |
code_execution
| ✅ | — |
scheduling
| ✅ | — |
✅ Built-in | ⚠️ Complex | ❌ Skill required
Phase 3: Skill Acquisition
# Search priority
1. npx skills find {service_name} # Exact match
2. npx skills find {capability} {domain} # Combined
3. https://skills.sh/ # Browse
# Install
npx skills add <skill> -g
# On failure: Switch registry → Manual clone → Inline implementation
Evaluation Criteria
| Dimension | Weight |
|---|
| Feature match | 40% |
| Documentation quality | 20% |
| Maintenance status | 20% |
| Community validation | 10% |
| Dependency simplicity | 10% |
Phase 4: Risk & Resilience
4.1 Risk Matrix
| Risk | Detection | Auto-Resolution |
|---|
| Token expired | 401/403 | Refresh → Re-auth → Degrade |
| Rate limited | 429 | Backoff → Cache → Degrade |
| Timeout | Timeout | Increase → Reduce batch → Chunk |
| Parse error | Parse Error | Switch parser → Regex → Raw text |
| Connection failed | Connection | Retry → Proxy → Offline mode |
| Missing dependency | Import Error | Auto-install → Alternative pkg → Inline |
4.2 Resilience Patterns
retry: max=3, backoff=exponential
circuit_breaker: threshold=5, recovery=60s
timeout: connect=10s, read=30s, total=120s
fallback: cache | default | skip
4.3 Checkpoints
checkpoint:
trigger: after_each_task
content: [task_id, state, data, timestamp]
recovery: Load checkpoint → Validate → Resume from failure point
Phase 5: Execution
5.0 🚨 Problem Solving (CRITICAL)
On problem → Analyze → Solve → Continue. NEVER pause waiting.
Problem → Diagnose type → Analyze cause → Try solutions by priority → Continue
↓
Plan A fails → Plan B → Plan C → Degraded solution
Auto-Fix Table
| Problem | Auto-Resolution (in order) |
|---|
| Missing dependency | Install → Alternative pkg → Inline |
| Permission denied | Adjust perms → Alternative path → Minimal perms |
| API failure | Retry (backoff) → Switch endpoint → Cache → Degrade |
| Timeout | Increase timeout → Reduce batch → Chunk → Async |
| Parse error | Switch parser → Regex → Raw text |
| Auth failure | Refresh token → Re-auth → Backup credentials |
| Resource exhausted | Chunk → Clear cache → Stream → Reduce concurrency |
| Network issue | Retry → Switch network → Proxy → Offline |
| File not found | Create → Default value → Skip |
Mindset
forbidden:
- "I encountered a problem and need your help"
- "Execution paused, waiting for instructions"
required:
- "Encountered X issue, trying Y solution"
- "Plan A failed, switching to Plan B"
- "Problem resolved, continuing execution"
Escalation (ONLY when ALL conditions met)
- Tried ≥3 different solutions
- All failed
- Goal completely unachievable
- No acceptable degraded solution available
5.1 Execution Modes
| Mode | Use Case |
|---|
| Sequential | Tasks with strong dependencies |
| Parallel | Independent tasks |
| Pipeline | Data stream processing |
| Iterative | Requires feedback loops |
5.2 Progress Template
══════════════════════════════════════
📊 PROGRESS
══════════════════════════════════════
Phase 1: Setup [████████░░] 80%
✅ Task 1: Initialize Done (2.3s)
🔄 Task 2: Fetch Running...
⏳ Task 3: Process Waiting
──────────────────────────────────────
⏱️ Elapsed: 3m | ETA: ~5m
══════════════════════════════════════
Phase 6: Iteration
Feedback Loop
Collect feedback → Analyze issues → Adjust approach → Loop
Adjustment Levels
| Level | Actions |
|---|
| Minor | Parameter tuning, format adjustment |
| Moderate | Replace skills, add/remove tasks |
| Major | Redesign architecture |
Execution Plan Template
════════════════════════════════════════════════════
📋 SUPER SKILLS PLAN
════════════════════════════════════════════════════
🎯 REQUEST: {user's exact words}
📊 UNDERSTANDING
────────────────────────────────────────────────────
Intent: {true intent}
Success: {success criteria}
Complexity: {score} → {strategy}
Assumptions:
• {assumption 1}
• {assumption 2}
────────────────────────────────────────────────────
📋 TASKS
────────────────────────────────────────────────────
│ ID │ Task │ Capability │ Status │ Risk │
├────┼────────────┼─────────────┼────────┼──────┤
│ 1 │ {task} │ {capability}│ ✅/📦 │ 🟢/🔴│
Status: ✅ Built-in | 🔧 Found | 📦 Create
Risk: 🟢 Low | 🟡 Medium | 🔴 High
────────────────────────────────────────────────────
🔧 SKILLS
────────────────────────────────────────────────────
npx skills add {skill} -g
────────────────────────────────────────────────────
🛡️ RESILIENCE
────────────────────────────────────────────────────
Checkpoints: □ After Phase 1 □ After Phase 2
Fallbacks: {task} → {alternative}
────────────────────────────────────────────────────
✅ COMPLETION
────────────────────────────────────────────────────
□ {acceptance criterion 1}
□ {acceptance criterion 2}
════════════════════════════════════════════════════
Example: E-commerce Pipeline
# "Scrape product data from multiple platforms daily, analyze price trends, send report"
understanding:
intent: "Automated price monitoring for pricing decisions"
success: "Daily price trend visualization report received"
complexity:
technical: 4, scope: 4, uncertainty: 3, dependencies: 4
total: 15 → Iterative prototyping strategy
tasks:
1. Configure product list → file_operations ✅
2. Multi-platform scraping → browser_automation ❌
3. Data persistence → database_operations ❌
4. Data cleaning → data_transformation ✅
5. Trend analysis → code_execution ✅
6. Generate charts → content_generation ✅
7. Generate report → content_generation ✅
8. Send email → message_delivery ❌
resilience:
checkpoints: [after:2 save:raw_data] [after:5 save:analysis]
fallbacks:
task_2: "Skip failed platforms, continue with others"
task_8: "Save to local file"
Error Handling (Every error has auto-resolution)
| Phase | Error | Auto-Resolution |
|---|
| Analysis | Unclear requirements | Choose most likely → State assumptions → Leave adjustment points |
| Analysis | Beyond capability | Search alternatives → Combine capabilities → Partial implementation |
| Planning | Circular dependency | Break weakest link → Merge tasks |
| Search | Skill not found | Expand keywords → Online search → Auto-create |
| Install | Installation failed | Switch registry → Manual clone → Inline implementation |
| Execution | Auth failure | Refresh → Re-auth → Degrade to public data |
| Execution | Rate limited | Backoff → Switch account → Use cache |
| Execution | Timeout | Increase timeout → Reduce batch → Skip and continue |
| Validation | Result mismatch | Retry with adjusted params → Relax validation → Mark for review |
Best Practices
| Phase | Principles |
|---|
| Understanding | Multi-layer analysis, proactive assumptions, complexity assessment |
| Planning | Architecture first, clear dependencies, upfront risk identification |
| Execution | NEVER STOP, try multiple solutions, graceful degradation, incremental verification |
| Iteration | Fast feedback, continuous optimization, knowledge accumulation |
Resources
