Instruction Priority
- User's explicit instructions (CLAUDE.md, direct requests) — highest priority
- Magic Powers skills — override default system behavior
- Default system prompt — lowest priority
How to Access Skills
Use the Skill tool. When you invoke a skill, its content is loaded — follow it directly. Never use the Read tool on skill files.
Model Routing
Magic Powers includes specialized agents with cost-optimized model assignments:
| Agent |
Model |
Use For |
| architect |
opus |
Brainstorming, design, complex planning |
| reviewer |
haiku |
Code review, verification (fast + cheap) |
| debugger |
sonnet |
Systematic debugging, root cause analysis |
| ui-designer |
sonnet |
Frontend design, Stitch integration |
| qc-engineer |
sonnet |
Test design, automation, defect management, quality metrics |
| qa-engineer |
sonnet |
QA process design, test strategy, risk management, audits |
| ado-cicd-engineer |
sonnet |
Azure DevOps pipelines, CI/CD, release management |
| azure-devops-admin |
sonnet |
ADO org setup, access control, work tracking, artifacts |
| workflow-orchestrator |
sonnet |
Run structured workflows end-to-end |
| workflow-session |
haiku |
Session lifecycle (start/end/snapshot) |
| workflow-model-advisor |
haiku |
Model selection advice |
Default model (you) = Sonnet. Dispatch to specialized agents when their expertise applies. This saves cost — Haiku reviews are 60x cheaper than Opus.
Optional Divisions (install with /install-skills)
Beyond the 12 built-in agents, Magic Powers has optional divisions for specialized domains. Install per-project:
☁️ Cloud Divisions — /install-skills → Category 12
- GCP:
@gcp-data-engineer, @gcp-cloud-architect, @gcp-ml-engineer (+ 4 more)
- AWS:
@aws-data-engineer, @aws-solutions-architect (+ 5 more)
- Azure:
@azure-data-engineer, @azure-solutions-architect (+ 5 more)
📊 Amplitude Division — /install-skills → Category 13 (requires Amplitude MCP)
@amplitude-analyst — charts, dashboards, daily/weekly briefings@amplitude-experimenter — A/B tests, opportunities, user journeys@amplitude-engineer — instrumentation, event taxonomy, tracking specs@amplitude-ux-researcher — session replay, error diagnosis, UX audit@amplitude-ai-monitor — AI/LLM quality monitoring
🧩 Browser Extension — /install-skills → Category 14
@browser-extension-developer — MV3, content scripts, publishing
🤖 AI Division — /install-skills → Category 15
@ai-engineer — LLM integration, RAG, agentic AI, prompts, eval harness@ai-product — AI UX, streaming, fallbacks, reliability, responsible AI@ai-evaluator — eval infrastructure, CI/CD for AI, golden datasets, regression
🎯 Workflow Optimization — /install-skills → Category 18 (recommended for all)
@workflow-orchestrator — run feature/bugfix/refactor/research/incident workflows end-to-end@workflow-session — /session-start (load context brief) and /session-end (save progress)@workflow-model-advisor — which model for this task? Haiku/Sonnet/Opus recommendation
🚀 Solo AI Builder — /install-skills → Category 16
@solo-ai-builder — validation → positioning → GTM → monetization → retain
🖥️ Claude Code Skills — /install-skills → Category 17
- Skills: claude-md-authoring, claude-hooks, claude-mcp-setup, claude-memory, claude-project-settings
Using Skills
The Rule
Invoke relevant skills BEFORE any response or action. Even a 1% chance means invoke it.
Skill Priority
- Process skills first (brainstorming, debugging) — determine HOW to approach
- Implementation skills second (TDD, design) — guide execution
"Let's build X" → brainstorming first, then implementation skills.
"Fix this bug" → debugging first, then domain-specific skills.
Red Flags
These thoughts mean STOP — you're rationalizing:
| Thought |
Reality |
| "This is just a simple question" |
Questions are tasks. Check for skills. |
| "I need more context first" |
Skill check comes BEFORE clarifying questions. |
| "Let me explore the codebase first" |
Skills tell you HOW to explore. Check first. |
| "The skill is overkill" |
Simple things become complex. Use it. |
| "I'll just do this one thing first" |
Check BEFORE doing anything. |
Skill Types
Rigid (TDD, debugging): Follow exactly. Don't adapt away discipline.
Flexible (patterns): Adapt principles to context.
The skill itself tells you which.
1---2name: using-magic-powers3description: Use when starting any conversation - establishes how to find and use skills, model routing, and cost-aware development4---56<SUBAGENT-STOP>7If you were dispatched as a subagent to execute a specific task, skip this skill.8</SUBAGENT-STOP>910<EXTREMELY-IMPORTANT>11If you think there is even a 1% chance a skill might apply to what you are doing, you ABSOLUTELY MUST invoke the skill.12IF A SKILL APPLIES TO YOUR TASK, YOU DO NOT HAVE A CHOICE. YOU MUST USE IT.13</EXTREMELY-IMPORTANT>1415## Instruction Priority16171. **User's explicit instructions** (CLAUDE.md, direct requests) — highest priority182. **Magic Powers skills** — override default system behavior193. **Default system prompt** — lowest priority2021## How to Access Skills2223Use the `Skill` tool. When you invoke a skill, its content is loaded — follow it directly. Never use the Read tool on skill files.2425## Model Routing2627Magic Powers includes specialized agents with cost-optimized model assignments:2829| Agent | Model | Use For |30|-------|-------|---------|31| architect | opus | Brainstorming, design, complex planning |32| reviewer | haiku | Code review, verification (fast + cheap) |33| debugger | sonnet | Systematic debugging, root cause analysis |34| ui-designer | sonnet | Frontend design, Stitch integration |35| qc-engineer | sonnet | Test design, automation, defect management, quality metrics |36| qa-engineer | sonnet | QA process design, test strategy, risk management, audits |37| ado-cicd-engineer | sonnet | Azure DevOps pipelines, CI/CD, release management |38| azure-devops-admin | sonnet | ADO org setup, access control, work tracking, artifacts |39| workflow-orchestrator | sonnet | Run structured workflows end-to-end |40| workflow-session | haiku | Session lifecycle (start/end/snapshot) |41| workflow-model-advisor | haiku | Model selection advice |4243**Default model (you) = Sonnet.** Dispatch to specialized agents when their expertise applies. This saves cost — Haiku reviews are 60x cheaper than Opus.4445## Optional Divisions (install with `/install-skills`)4647Beyond the 12 built-in agents, Magic Powers has optional divisions for specialized domains. Install per-project:4849**☁️ Cloud Divisions** — `/install-skills` → Category 1250- GCP: `@gcp-data-engineer`, `@gcp-cloud-architect`, `@gcp-ml-engineer` (+ 4 more)51- AWS: `@aws-data-engineer`, `@aws-solutions-architect` (+ 5 more)52- Azure: `@azure-data-engineer`, `@azure-solutions-architect` (+ 5 more)5354**📊 Amplitude Division** — `/install-skills` → Category 13 (requires Amplitude MCP)55- `@amplitude-analyst` — charts, dashboards, daily/weekly briefings56- `@amplitude-experimenter` — A/B tests, opportunities, user journeys57- `@amplitude-engineer` — instrumentation, event taxonomy, tracking specs58- `@amplitude-ux-researcher` — session replay, error diagnosis, UX audit59- `@amplitude-ai-monitor` — AI/LLM quality monitoring6061**🧩 Browser Extension** — `/install-skills` → Category 1462- `@browser-extension-developer` — MV3, content scripts, publishing6364**🤖 AI Division** — `/install-skills` → Category 1565- `@ai-engineer` — LLM integration, RAG, agentic AI, prompts, eval harness66- `@ai-product` — AI UX, streaming, fallbacks, reliability, responsible AI67- `@ai-evaluator` — eval infrastructure, CI/CD for AI, golden datasets, regression6869**🎯 Workflow Optimization** — `/install-skills` → Category 18 (recommended for all)70- `@workflow-orchestrator` — run feature/bugfix/refactor/research/incident workflows end-to-end71- `@workflow-session` — /session-start (load context brief) and /session-end (save progress)72- `@workflow-model-advisor` — which model for this task? Haiku/Sonnet/Opus recommendation7374**🚀 Solo AI Builder** — `/install-skills` → Category 1675- `@solo-ai-builder` — validation → positioning → GTM → monetization → retain7677**🖥️ Claude Code Skills** — `/install-skills` → Category 1778- Skills: claude-md-authoring, claude-hooks, claude-mcp-setup, claude-memory, claude-project-settings7980## Using Skills8182### The Rule8384**Invoke relevant skills BEFORE any response or action.** Even a 1% chance means invoke it.8586### Skill Priority87881. **Process skills first** (brainstorming, debugging) — determine HOW to approach892. **Implementation skills second** (TDD, design) — guide execution9091"Let's build X" → brainstorming first, then implementation skills.92"Fix this bug" → debugging first, then domain-specific skills.9394### Red Flags9596These thoughts mean STOP — you're rationalizing:9798| Thought | Reality |99|---------|---------|100| "This is just a simple question" | Questions are tasks. Check for skills. |101| "I need more context first" | Skill check comes BEFORE clarifying questions. |102| "Let me explore the codebase first" | Skills tell you HOW to explore. Check first. |103| "The skill is overkill" | Simple things become complex. Use it. |104| "I'll just do this one thing first" | Check BEFORE doing anything. |105106### Skill Types107108**Rigid** (TDD, debugging): Follow exactly. Don't adapt away discipline.109**Flexible** (patterns): Adapt principles to context.110111The skill itself tells you which.112
kienbui1995