claude-system
PROACTIVELY consult Claude Code for creative design, ideation, and ambiguous requirement exploration. Claude Code excels at: creative problem-solving, architectural ideation, interactive dialogue for unclear requirements, and fast file operations. Use when you need fresh perspectives, design brainstorming, or help clarifying vague user requests. Explicit triggers: "design ideas", "brainstorm", "explore requirements", "clarify".
How do I install this agent skill?
npx skills add https://github.com/jey3dayo/apm-workspace --skill claude-systemIs this agent skill safe to install?
- Gen Agent Trust Hubfail
The skill integrates with the Claude Code CLI but recommends high-risk practices like bypassing security permissions and using unvalidated user input in shell commands.
- Socketwarn
1 alert: gptAnomaly
- Snykwarn
Risk: MEDIUM · 1 issue
What does this agent skill do?
Claude System Integration
Claude Code is your creative design partner with exceptional ideation and architectural thinking abilities
Context Management (CRITICAL)
Use Claude Code with awareness of context consumption
| Situation | Recommended approach |
|---|---|
| Short questions / short answers | Direct invocation OK |
| Architecture design consultation | Via subprocess (interactive exploration) |
| Requirements clarification / ambiguity | Via subprocess (deep dialogue) |
| File operations | Via subprocess |
┌──────────────────────────────────────────────────────────┐
│ Codex / Gemini │
│ → For short questions, direct invocation is OK │
│ → For interactive exploration, use subprocess │
│ │
│ ┌────────────────────────────────────────────────────┐ │
│ │ Subprocess (background) │ │
│ │ → Calls Claude Code │ │
│ │ → Interactive design exploration │ │
│ │ → Saves design artifacts to .claude/docs/design/ │ │
│ │ → Returns design decisions and key insights │ │
│ └────────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────────┘
About Claude Code
Claude Code excels at:
- Creative problem-solving — Novel solutions and fresh perspectives
- Architectural ideation — System design and component structure brainstorming
- Interactive dialogue — Clarifying unclear requirements through conversation
- Fast file operations — Read/Write/Edit operations with specialized tools
Think of Claude Code as your design partner who helps you think through complex problems creatively.
When you need creative thinking → Use subprocess → Subprocess consults Claude Code
Claude vs Codex vs Gemini: Choose the Right Tool
| Task | Claude | Codex | Gemini |
|---|---|---|---|
| Creative design ideation | ✓ | ||
| Architectural brainstorming | ✓ | ||
| Clarifying vague requirements | ✓ | ||
| Fast file operations | ✓ | ||
| Reasoning & analysis | ✓ | ||
| Code implementation | ✓ | ||
| Debugging | ✓ | ||
| Large codebase research | ✓ | ||
| Multimodal processing | ✓ | ||
| Latest docs/library research | ✓ |
When to Consult Claude Code
ALWAYS consult Claude Code BEFORE:
- Creative design & ideation - New feature concepts, architectural patterns
- Ambiguous requirements - When user's intent is unclear or underspecified
- Design exploration - Exploring multiple approaches, brainstorming solutions
- Fast file operations - Bulk read/write operations, rapid prototyping
Trigger Phrases (User Input)
Consult Claude Code when user says:
| Japanese (triggers) | English equivalent |
|---|---|
| 「デザインアイデアを出して」「設計して」 | "Design ideas" "Design this" |
| 「ブレストして」「アイデア出して」 | "Brainstorm" "Generate ideas" |
| 「要件を整理して」「明確にして」 | "Clarify requirements" "Explore" |
| 「アーキテクチャを考えて」 | "Architect this" "Design structure" |
| 「ファイルを一括で編集」 | "Bulk edit files" |
When NOT to Consult
Skip Claude Code for:
- Deep reasoning (use Codex instead)
- Code debugging (use Codex instead)
- Research tasks (use Gemini instead)
- Multimodal processing (use Gemini instead)
- Straightforward implementation (use Codex instead)
How to Consult (via Subprocess)
IMPORTANT: Use subprocess to preserve main context
Recommended: Subprocess Pattern
Create a background subprocess that calls Claude Code:
# Design exploration pattern
(
claude -p "Design question or task" \
--allowedTools "Read,Write,Edit,Glob,Grep,Bash" \
--model sonnet \
2>/dev/null > /tmp/claude_output.txt
# Save design artifacts
mkdir -p .claude/docs/design/
cp /tmp/claude_output.txt .claude/docs/design/design-exploration.md
# Return summary
echo "=== Design Summary ==="
head -n 20 /tmp/claude_output.txt
) &
Subprocess Patterns by Task Type
Design Ideation Pattern
claude -p "Generate 3-5 design alternatives for {feature}.
Include pros/cons and architectural implications." \
--allowedTools "Read,Glob,Grep" \
--model sonnet \
2>/dev/null > .claude/docs/design/design-alternatives.md
Requirement Clarification Pattern
claude -p "The user requested: '{vague_request}'.
Explore possible interpretations and create clarifying questions." \
--allowedTools "Read,Glob,Grep" \
--model sonnet \
2>/dev/null > .claude/docs/design/requirement-clarification.md
Bulk File Operation Pattern
claude -p "Bulk operation task: {description}
Files to process: {file_list}
Operation: {operation_details}" \
--allowedTools "Read,Write,Edit,Glob,Grep,Bash" \
--model sonnet \
2>/dev/null > .claude/docs/design/bulk-operation-log.md
Architectural Design Pattern
claude -p "Design architecture for {system}.
Consider: scalability, maintainability, testability.
Output: component diagram (text), module structure, interaction patterns." \
--allowedTools "Read,Write,Edit,Glob,Grep" \
--model sonnet \
2>/dev/null > .claude/docs/design/architecture.md
Step 2: Monitor Progress
Check subprocess output:
# Check if still running
ps aux | grep claude
# Monitor output in real-time
tail -f .claude/docs/design/design-exploration.md
Step 3: Receive Design Artifacts
Claude Code saves design artifacts to .claude/docs/design/. Extract key insights and report to user in Japanese.
Claude CLI Commands Reference
For use within subprocesses:
# Creative design exploration (read-only)
claude -p "{design question}" \
--allowedTools "Read,Glob,Grep" \
--model sonnet \
2>/dev/null
# Design with implementation (write enabled)
claude -p "{design and implement task}" \
--allowedTools "Read,Write,Edit,Glob,Grep,Bash" \
--model sonnet \
2>/dev/null
# Fast iteration (Haiku for speed)
claude -p "{quick design question}" \
--allowedTools "Read,Glob,Grep" \
--model haiku \
2>/dev/null
# Complex design (Opus for depth)
claude -p "{complex architectural design}" \
--allowedTools "Read,Write,Edit,Glob,Grep,Bash" \
--model opus \
2>/dev/null
# Sandboxed environment (skip permissions)
claude -p "{task}" \
--dangerously-skip-permissions \
--allowedTools "Read,Write,Edit,Bash" \
2>/dev/null
Key CLI Options
| Option | Purpose |
|---|---|
-p, --print | Non-interactive mode (required) |
--allowedTools | Restrict tool access |
--model sonnet|opus|haiku | Model selection |
--system-prompt | Custom system prompt |
--max-budget-usd | API cost limit |
--dangerously-skip-permissions | Skip permission prompts (sandbox) |
Language Protocol
- Ask Claude in English
- Subprocess receives response in English or Japanese (Claude respects user's CLAUDE.md)
- Subprocess saves full output to
.claude/docs/design/ - Main receives summary, reports to user in Japanese
Why Subprocess Pattern?
- Context preservation: Main orchestrator stays lightweight
- Full capture: Subprocess can save entire Claude output to file
- Concise handoff: Main only receives key insights
- Parallel work: Background subprocesses enable concurrent design exploration
Task Templates
Template 1: Design Exploration
claude -p "Design {feature/system}.
Requirements: {requirements}
Constraints: {constraints}
Output: 3-5 design alternatives with trade-off analysis." \
--allowedTools "Read,Glob,Grep,Write" \
--model sonnet \
2>/dev/null > .claude/docs/design/design-exploration-$(date +%Y%m%d-%H%M%S).md
Template 2: Requirement Clarification
claude -p "User request: '{user_request}'
Current codebase context: {context}
Task: Identify ambiguities and generate clarifying questions." \
--allowedTools "Read,Glob,Grep" \
--model sonnet \
2>/dev/null > .claude/docs/design/requirement-clarification-$(date +%Y%m%d-%H%M%S).md
Template 3: Architectural Brainstorming
claude -p "Brainstorm architectural approaches for {system}.
Current stack: {tech_stack}
Key requirements: {requirements}
Output: Component structure, interaction patterns, technology choices." \
--allowedTools "Read,Glob,Grep,Write" \
--model opus \
2>/dev/null > .claude/docs/design/architecture-brainstorm-$(date +%Y%m%d-%H%M%S).md
Template 4: Bulk File Operation
claude -p "Bulk edit task: {description}
Target files: {file_pattern}
Operation: {edit_description}
Safety: Preview changes before applying." \
--allowedTools "Read,Write,Edit,Glob,Grep,Bash" \
--model sonnet \
2>/dev/null > .claude/docs/design/bulk-operation-$(date +%Y%m%d-%H%M%S).md
Integration Best Practices
- Clear delegation boundaries: Use Claude for design, Codex for reasoning, Gemini for research
- Structured output: Save design artifacts to
.claude/docs/design/for future reference - Subprocess isolation: Run Claude in background to avoid context contamination
- Cost awareness: Use Haiku for quick tasks, Sonnet for standard, Opus for complex
- Tool restrictions: Limit
--allowedToolsto minimum required for task
Error Handling
If Claude CLI fails:
# Check authentication
claude --version
# Check Claude Code installation
which claude
# Verify permissions
ls -la ~/.claude/
# Test basic invocation
claude -p "test" --model sonnet 2>&1 | head
Use Claude Code (via subprocess) for creative design, Codex for reasoning, Gemini for research
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/jey3dayo/apm-workspace/claude-system">View claude-system on skillZs</a>