메뉴
Load required context before any work. Uses prompt matrix to determine what to load based on request type. Prevents premature implementation and context poisoning.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.
Systematically identify what's missing, unknown, or needs investigation before making changes or decisions. Use proactively when proposing features, implementing changes, or making technical decisions to ensure all critical areas are covered. This skill catches overlooked dependencies, edge cases, and integration points before they become problems. Examples: - "Implement user authentication" → trigger gap analysis (what's the current auth system?) - "Add caching to API" → trigger gap analysis (what caching infrastructure exists?) - "Migrate to TypeScript" → trigger gap analysis (what's the build pipeline?) - "Refactor this module" → trigger gap analysis (what depends on this module?) - "Add new API endpoint" → trigger gap analysis (what endpoints exist, what patterns?)
Detect ambiguous or underspecified research requests and guide through structured clarification questions before executing any research. Use proactively when user requests are vague, lack specificity, missing context, or have unclear goals. This skill prevents wasted research by ensuring intent is crystal clear before any web searches, file reads, or data gathering. Examples: - "Research the codebase" → ❌ Too vague, trigger clarification - "Why is this slow?" → ❌ Missing context (what is "this"?), trigger clarification - "Add a feature" → ❌ Not research, this is implementation - "Investigate authentication flow" → ❌ What specifically about it? Trigger clarification - "Compare React and Vue" → ✅ Specific enough, but still validate dimensions
Evaluate requests, proposals, or decisions from multiple dimensions (technical, business, UX, security, performance, maintainability, etc.) to ensure holistic understanding before proceeding. Use proactively when making technical decisions, choosing solutions, or proposing changes. This skill prevents single-minded thinking and uncovers trade-offs, risks, and opportunities across all relevant aspects. Examples: - "Use Redis for caching" → assess multiple aspects (performance, cost, complexity, ops) - "Add AI to our app" → assess aspects (technical feasibility, user experience, data privacy) - "Refactor to microservices" → assess aspects (benefits, costs, migration complexity, team readiness) - "Implement authentication" → assess aspects (security, UX, integration effort, maintenance) - "Switch to React 19" → assess aspects (benefits, migration cost, breaking changes, learning curve)
Choose the right OpenCode primitive (Command, Skill, Tool, Prompt, CustomTool, Plugin hook) for a given task or workflow. Use proactively when deciding which OpenCode capability to use, or when trying to understand what's available. This skill maps research and development needs to appropriate OpenCode mechanisms, ensuring optimal use of the platform's capabilities. Examples: - "How do I search the codebase?" → select tool (read/grep) - "I need a reusable workflow for feature development" → select command - "How do I intercept file writes?" → select plugin hook - "What primitive should I use for systematic debugging?" → select skill - "I need a custom tool for X" → select CustomTool (if innate tools insufficient)
Create structured, executable research plans with clear phases, checkpoints, and validation criteria. Use proactively when research intent is clear but execution plan is missing. This skill transforms research goals into actionable workflows with defined deliverables, time estimates, and success metrics, ensuring research is systematic and produces useful outputs. Examples: - "Research how to add OAuth" (intent clear, need plan) → trigger workflow planner - "Investigate performance bottlenecks" (intent clear, need plan) → trigger workflow planner - "Analyze codebase architecture" (intent clear, need plan) → trigger workflow planner - "Find best practices for X" (intent clear, need plan) → trigger workflow planner - "Compare technology options" (intent clear, need plan) → trigger workflow planner
| name | context-first |
| description | Load required context before any work. Uses prompt matrix to determine what to load based on request type. Prevents premature implementation and context poisoning. |
TRAP 1 & 2 Defense: Load context before implementation
Never implement before understanding. Context loading is mandatory before any code changes.
Before ANY implementation:
Read existing code to understand:
Load artifacts based on request type:
| Request Category | Context to Load |
|---|---|
| New feature | Product requirements, UX specs |
| Feature extension | Architecture, existing patterns |
| Bug fix | Sprint status, story context |
| Refactoring | Architecture, project truths |
| Documentation | Architecture, existing docs |
| Ambiguous | AGENTS.md, project truths |
Confirm approach before coding:
Documents have different lifespans:
| Document Type | Typical Lifespan | Refresh Trigger |
|---|---|---|
| Architecture | Long-lived | Major changes only |
| ADRs | Long-lived | Superseded by new ADR |
| Stories | Sprint-lived | Sprint completion |
| Tech specs | Story-lived | Story completion |
| Plans | Task-lived | Task completion |
Check modification timestamps. Stale artifacts (> 2 hours for short-lived, > 24 hours for long-lived) may need refresh.
Documents iterate on each other in tiers:
Tier 1: Governing (Architecture, PRD, AGENTS.md)
↓ informs
Tier 2: Planning (Epics, Stories, ADRs)
↓ informs
Tier 3: Execution (Tech Specs, Tasks)
↓ informs
Tier 4: Ephemeral (Session notes, scratch files)
Always validate lower tiers against higher tiers. If Tier 3 conflicts with Tier 1, Tier 1 wins.
Use project context loader tools to load based on request classification:
If validation fails:
⛔ CONTEXT FIRST VIOLATION
Attempted: {action}
Missing: {missing_context}
Tier conflict: {if any}
ACTION: Complete 3-Step Validation before proceeding.