Anthropic's AI-powered terminal coding assistant with MCP Server integration.
# Install Node
nvm install node && nvm alias default node
# Install uv (for Python-based MCP servers)
brew install uvnpm install -g @anthropic-ai/claude-code
# Environment variables for non-English input lag fix
echo 'export TERM=xterm-256color' >> ~/.bashrc
echo 'export LC_ALL=C.UTF-8' >> ~/.bashrcclaude
> /loginexport AWS_ACCESS_KEY_ID={your-key}
export AWS_SECRET_ACCESS_KEY={your-secret}
export AWS_REGION_NAME=us-west-1
export CLAUDE_CODE_USE_BEDROCK=1export ANTHROPIC_BASE_URL={your-gateway-url}
export ANTHROPIC_AUTH_TOKEN={your-auth-token}claude # New session
claude -c # Continue last session
claude -r # Resume specific session> /clear # Reset context
> /model opus # Switch to Opus 4.5
> /model sonnet # Switch to Sonnet 4.5
> /model sonnet[1m] # Sonnet 4.5 with 1M context
> @{file-path} {prompt} # Include files
> {prompt} ultrathink # Enable extended thinking
> /compact "{instructions}" # Manual context compactionMode Switching:
TAB- Toggle extended thinkingSHIFT+TAB(2x) - Enter Plan Mode (read-only analysis)SHIFT+TAB(1x) - Exit to Edit Mode
claude mcp add time -s user -- uvx mcp-server-timeclaude mcp add context7 -s user -- npx -y @upstash/context7-mcpGet API key from brave.com/search/api (Free: 5,000 queries/month)
claude mcp add-json -s user brave-search '{"command":"npx","args":["-y","brave-search-mcp"],"env":{"BRAVE_API_KEY":"{your-key}"}}'claude mcp add fetch -s user -- uvx mcp-server-fetchclaude mcp add reddit -s user -- uvx --from git+https://github.com/adhikasp/mcp-reddit.git mcp-redditnpm install -g @executeautomation/playwright-mcp-server
claude mcp add playwright -s user -- npx -y @executeautomation/playwright-mcp-server# Run from project root
claude mcp add serena -- uvx --from git+https://github.com/oraios/serena serena start-mcp-server --context ide-assistant --project $(pwd)
# Initial onboarding
claude
> start Serena onboardingclaude mcp add-json -s user slack '{"command":"npx","args":["-y","slack-mcp-server@latest"],"env":{"SLACK_MCP_XOXP_TOKEN":"{your-token}"}}'claude mcp add-json -s user notion '{"command":"npx","args":["-y","@notionhq/notion-mcp-server"],"env":{"NOTION_TOKEN":"{your-token}"}}'Claude Code uses CLAUDE.md files as persistent memory:
| Path | Scope |
|---|---|
~/.claude/CLAUDE.md |
Global (all projects) |
./CLAUDE.md |
Project-specific |
The CLAUDE.md in this repository root serves as the global memory template, defining behavioral rules, MCP server usage patterns, and response preferences.
This repository is a Claude Code Plugin Marketplace containing the deep-thinking plugin.
# Add marketplace (one-time setup)
/plugin marketplace add JSON-OBJECT/claude-code
# Install plugin
/plugin install deep-thinking@jsonobject-marketplace
# Restart Claude Code to load the pluginBefore you research, know WHAT to research.
Scans 5-8 subreddits × 75+ posts to identify what's genuinely hot RIGHT NOW.
| Feature | What It Does |
|---|---|
| Heat Score Formula | Objective ranking: (Upvotes × 0.4) + (Comments × 0.3) + (Cross-sub × 0.2) + (Recency × 0.1) |
| News Cross-Validation | Compares Reddit trends against mainstream news — prevents echo chamber reports |
| Emerging Signals Watch | Tracks low-heat topics with rising momentum — tomorrow's headlines today |
| Actionable Output | Every issue includes a ready-to-run /deep-research query |
Use before: /deep-research — ensures you research the RIGHT topic, not just any topic.
One-shot research that answers questions before you ask them.
Conducts 15-20+ searches with mandatory Phase Zero "Unknown Unknowns" discovery.
| Feature | What It Does |
|---|---|
| Phase Zero Protocol | Questions the question itself — catches terminology confusion, outdated assumptions, missed prerequisites |
| One-Shot Completeness | Anticipates 5 follow-up questions ("What is it?" → "How much?" → "Gotchas?" → "Show code" → "Verdict?") and answers ALL in one report |
| Ki-Sho-Ten-Ketsu Structure | Four-act narrative (Setup → Development → Turn → Conclusion) — reads like investigative journalism, not bullet lists |
| Community + Official Balance | Reddit/HN sentiment weighted proportionally against official sources — 70% positive / 30% concern = both reported |
Output: Publication-ready deep dive that feels like Ars Technica meets Gemini Deep Research.
Build prompts that resist rationalization.
Creates bulletproof instructions following the Superpowers philosophy with 9 mandatory components.
| Component | Purpose |
|---|---|
| Iron Law | ONE non-negotiable rule that guarantees failure if broken |
| Red Flags Section | Mental patterns that signal you're about to fail — "If you catch yourself thinking..." |
| Rationalization Table | Preempts every excuse with direct rebuttal — Excuse | Reality format |
| Phase Gates | Clear checkpoints that MUST be passed before proceeding |
| Strong Language Protocol | MUST/NEVER/ALWAYS — no soft language ("should", "consider", "try to") |
Philosophy: If you think you don't need the structure, you need it most.
Meeting notes that tell a story, not list fragments.
Transforms raw transcripts into narrative-driven documentation with 6-phase verification.
| Feature | What It Does |
|---|---|
| Counterparty Intelligence | External meetings: 3+ searches per counterparty for context (size, tech stack, recent news) |
| STT Error Correction | Catches phonetic mishearings ("jenkinsun" → "Jenkins") with 2+ source verification |
| Immediate Inline Definitions | Technical terms explained right where they appear (> 📘 format), not buried in glossary |
| Storytelling Tone + Metaphors | Mandatory 1-2 metaphors per document — "From gatekeeper to guardrail" style |
| Zero Detail Loss | Every business-relevant point preserved — 400-5000+ words scaled to source density |
Anti-pattern: Bullet lists. This produces flowing narrative with cause-and-effect reasoning.
Rewriting in Korean, not converting English structures.
Transcreates English IT articles into native Korean with terminology verification and 10 anti-artifact rules.
| Feature | What It Does |
|---|---|
| Tier 1/2 Verification | Established terms: 1 official source / New terms: 2+ sources including community (velog, tistory) |
| 10 Anti-Translation-Artifact Rules | Pronoun elimination, Passive→Active, Sino-Korean ban, Emotional modifier ban, etc. |
| Community Standard Priority | If Korean dev community uses English 70%+ of the time, keep English (Token, Prompt, Fine-tuning) |
| Code Block Protection | NEVER modifies content inside code blocks or inline code |
Key insight: "Fine-tuning" → "파인튜닝" (community usage), NOT "미세 조정" (textbook translation).
Anti-AI-looking blog covers that make readers curious.
Generates image prompts for Gemini Nano Banana Pro with 6-phase visual storytelling.
| Feature | What It Does |
|---|---|
| Visual Metaphor First | Every cover starts with an unexpected analogy — "Kubernetes = postal worker sorting containers" |
| Chain-of-Thought Composition | Explicit 9-grid spatial positioning — asymmetry always, center composition = AI tell |
| Anti-AI Armor | Mandatory "Avoid" section with 8+ items (symmetry, HDR, plastic textures, generic tech imagery) |
| Imperfection Keywords | 2-3 required: "visible brushstrokes", "analog film grain", "slightly uneven lines" |
| Title Typography Mandatory | Every prompt includes title overlay specifications — covers without title = incomplete |
Artist references: Christoph Niemann, Saul Steinberg, The New Yorker editorial style.
| Command | Required | Optional |
|---|---|---|
/deep-thinking:pulse |
Time, Brave Search, Reddit | - |
/deep-thinking:deep-research |
Time, Brave Search, Reddit | - |
/deep-thinking:forge-prompt |
None | - |
/deep-thinking:meeting-notes |
Time, Brave Search | Context7, Fetch |
/deep-thinking:translate-kr |
Brave Search | - |
/deep-thinking:blog-cover |
None | - |