Skip to main content

Claude Code

Setup​

  1. Sync config with cc-switch.
sqlite3 ~/.cc-switch/cc-switch.db "DELETE FROM session_log_sync;"
sqlite3 ~/.cc-switch/cc-switch.db "DELETE FROM skill_repos;"
sqlite3 ~/.cc-switch/cc-switch.db "DELETE FROM skills;"
  1. Set up .claude.json:
sed -i '0,/{/s/{/{\n  "hasCompletedOnboarding": true,/' ~/.claude.json
  1. Set up harness:
claude plugin marketplace add anthropics/claude-plugins-official
claude plugin marketplace add jarrodwatts/claude-hud
claude plugin marketplace add sabertaz/skills
claude plugin install ralph-loop@claude-plugins-official
claude plugin install superpowers@claude-plugins-official
claude plugin install claude-hud@claude-hud
claude plugin install sabertaz@sabertaz
pnpx skills add anthropics/skills -s skill-creator -a claude-code codex opencode -g -y
pnpx skills add mattpocock/skills -s grill-me grill-with-docs -a claude-code codex opencode -g -y
pnpx skills add vercel-labs/skills -s find-skills -a claude-code codex opencode -g -y
pnpx skills add vercel-labs/agent-skills -s "*" -a claude-code codex opencode -g -y
pnpx skills add multica-ai/andrej-karpathy-skills -s "*" -a claude-code codex opencode -g -y
pnpx skills add nextlevelbuilder/ui-ux-pro-max-skill -s "*" -a claude-code codex opencode -g -y
pnpx skills add upstash/context7 -s find-docs -a claude-code codex opencode -g -y
Update
claude plugin update ralph-loop@claude-plugins-official
claude plugin update superpowers@claude-plugins-official
claude plugin update claude-hud@claude-hud
claude plugin update sabertaz@sabertaz
pnpx skills update
Proxy

Optionally set up cli-proxy-api for API proxying.

Keyboard Shortcuts​

Cheat sheet:

  • <C-R>: search history.
  • <C-G>: open prompt in editor.

Memory​

Captures everything Claude does during your coding sessions, compresses it with Claude agent-sdk, and injects relevant context back into future sessions.

Session​

Resume from session:

# Most recent session
claude -c

# Specific session
claude -r
claude -r [session_id]

Noninteractive Mode​

Use Claude Code as a Unix-style utility:

claude -p "$REVIEW_PROMPT" > review-result.md 2>&1 || true

See GitHub code review action, improved version, and agent-native version.

Composable

Unix 系统中, 命令行工具通常是独立的、可组合的， claude -p 可以接受标准输入, 并且可以与其他工具无缝集成, 从而形成一个强大的工具链.

Commands​

.claude/commands/api/api-new.md:

Create a new POST $ARGUMENTS endpoint with:

- Auth middleware.
- Rate limiting middleware.
- Error handling wrapper.
- TypeScript interfaces for request and response.
/api-new /api/users

URL to Markdown command.

Skills​

<skills_instructions>
When users ask you to perform tasks, check if any of the available skills below can help complete the task more effectively.

How to use skills:
- Invoke skills using this tool with the skill name only (no arguments)
- When you invoke a skill, you will see <command-message>The "{name}" skill is loading</command-message>
- The skill's prompt will expand and provide detailed instructions

IMPORTANT:
- Only use skills listed in <available_skills> below
- Do not invoke a skill that is already running
</skills_instructions>

<available_skills>
<skill>
<name>xlsx</name>
<description>Comprehensive spreadsheet creation, editing, and analysis with support for formulas, formatting, data analysis...</description>
<location>plugin</location>
</skill>
</available_skills>

渐进式披露:

  • 核心功能简洁: 500 行以内.
  • 高级功能分离.
  • 按需加载: metadata -> instructions -> resources and code.
  • Persistent storage + on-demand loading + human-editable.
工具选择
  • 临时任务 -> prompts
  • 标准作业程序 -> skills
  • 连接外部系统 -> MCP
  • 复杂独立任务 -> subagents

Name​

Using gerund form verb + -ing: this clearly describes activity or capability:

  • processing-pdf
  • analyzing-spreadsheets
  • managing-databases
  • testing-code
  • writing-documentation

Description​

核心原则: 既要说明做什么，也要说明什么时候用:

description: Extract text and tables from PDF files, fill forms, merge documents.
  Use when working with PDF files or when the user mentions PDFs,
  forms, or document extraction.

Patterns​

Common skills patterns:

  • Sequential workflow orchestration.
  • Multi-MCP coordination.
  • Iterative refinement: ralph loop, etc.
  • Context-aware tool selection.
  • Domain-specific intelligence: finance, marketing, math, etc.

Effective​

  1. Core quality:
    • Description is specific and includes key terms
    • Description includes both what the Skill does and when to use it
    • SKILL.md body is under 500 lines
    • Additional details are in separate files (if needed)
    • No time-sensitive information (or in "old patterns" section)
    • Consistent terminology throughout
    • Examples are concrete, not abstract
    • File references are one level deep
    • Progressive disclosure used appropriately
    • Workflows have clear steps
  2. Code and scripts:
    • Scripts solve problems rather than punt to Claude
    • Error handling is explicit and helpful
    • No "voodoo constants" (all values justified)
    • Required packages listed in instructions and verified as available
    • Scripts have clear documentation
    • No Windows-style paths (all forward slashes)
    • Validation/verification steps for critical operations
    • Feedback loops included for quality-critical tasks
  3. Testing:
    • At least three evaluations created
    • Tested with Haiku, Sonnet, and Opus
    • Tested with real usage scenarios
    • Team feedback incorporated (if applicable)

Security​

  1. Code Review:
    • 所有脚本已审查
    • 无硬编码密钥或密码
    • 无危险的系统命令（rm -rf, eval, exec）
    • 文件路径经过验证（防止路径遍历）
  2. Network Access:
    • 检查所有外部 URL
    • 验证 API 端点可信
    • 处理网络失败情况
  3. Data Handling:
    • 无敏感数据泄露
    • 日志不包含 PII
    • 临时文件正确清理
  4. Permissions:
    • 最小权限原则
    • 不请求不必要的文件访问
    • 明确说明需要的权限
  5. Documentation:
    • 安全注意事项已文档化
    • 数据处理流程透明
    • 用户知情同意
grep -r "requests\." .
grep -r "os.system" .
grep -r "subprocess" .
grep -r "eval" .

Loops​

Loops are agents repeating cycles of work until a stop condition is met:

  • /goal: Iterate until verifiable condition or turn cap is reached. Deterministic criteria (e.g. tests passed, score threshold) work best, generalizes the DIY Ralph loop. 
    /goal get the homepage Lighthouse score to 90 or above, stop after 5 tries
  • /loop: Rerun prompt on local interval, stops when your machine is off. 
    /loop 5m check my PR, address review comments, and fix failing CI
  • /schedule: Move recurring routine to the cloud. Combine with /goal, skills, and auto mode for proactive loops (e.g. triage, migrations, dependency upgrades).

Plugins​

Install plugins from marketplace, it will extend Claude Code through the plugin system with:

  • Custom commands.
  • Agents.
  • Hooks.
  • Skills.
  • MCP servers.

SDK​

Proxy​

  1. Set up cli-proxy-api:
yay -S cli-proxy-api-bin
mkdir -p ~/.cli-proxy-api
cp /usr/share/doc/cli-proxy-api-bin/config.example.yaml ~/.cli-proxy-api/config.yaml
port: 8317

remote-management:
  secret-key: MGT-123456

api-keys:
  - sk-dummy

usage-statistics-enabled: true

  1. Set up providers with Web UI.

  2. Set up service:

systemctl --user enable --now cli-proxy-api.service

Best Practices​

Claude code best practices:

  • Research mode: deeply/intricacies/in great details/go through everything.
  • Plan mode: 对于复杂的任务, 使用 plan mode 先进行规划, 会比直接执行更高效.
  • Extended thinking mode: MAX_THINKING_TOKENS.
  • Claude code has built-in toolkit, follow certain paradigm to use them.
  • Avoid /compact: dump plan and progress + /clear the state + restart from dump file.
  • Use GitHub actions: code review, issue triage, documentation sync, etc.

Troubleshooting​

Bug Release​

Remove ~/.claude/projects/ for bug release.

Official Marketplace​

src/hooks/useOfficialMarketplaceNotification.tsx output Failed to install Anthropic marketplace · Will retry on next startup, src/utils/plugins/officialMarketplaceStartupCheck.ts checks:

  • officialMarketplaceAutoInstallFailReason
  • officialMarketplaceAutoInstallRetryCount
  • officialMarketplaceAutoInstallNextRetryTime

References​