AI Agent Workflow (Workflow & Productivity)
When to use this skill
- Optimize everyday AI agent work
- Integrate Git/GitHub workflows
- Use MCP servers
- Manage and recover sessions
- Apply productivity techniques
1. Key commands by agent
Claude Code commands
| Command | Function | When to use |
|---|
/init | Auto-generate a CLAUDE.md draft | Start a new project |
/usage | Show token usage/reset time | Start of every session |
/clear | Clear conversation history | When context is polluted; start a new task |
/context | Context window X-Ray | When performance degrades |
/clone | Clone the entire conversation | A/B experiments; backups |
/mcp | Manage MCP servers | Enable/disable MCP |
!cmd | Run immediately without Claude processing | Quick status checks |
Gemini CLI commands
| Command | Function |
|---|
gemini | Start a conversation |
@file | Add file context |
-m model | Select model |
Codex CLI commands
| Command | Function |
|---|
codex | Start a conversation |
codex run | Run a command |
2. Keyboard shortcuts (Claude Code)
Essential shortcuts
| Shortcut | Function | Importance |
|---|
Esc Esc | Cancel the last task immediately | Highest |
Ctrl+R | Search prompt history | High |
Shift+Tab x2 | Toggle plan mode | High |
Tab / Enter | Accept prompt suggestion | Medium |
Ctrl+B | Send to background | Medium |
Ctrl+G | Edit in external editor | Low |
Editor editing shortcuts
| Shortcut | Function |
|---|
Ctrl+A | Move to start of line |
Ctrl+E | Move to end of line |
Ctrl+W | Delete previous word |
Ctrl+U | Delete to start of line |
Ctrl+K | Delete to end of line |
3. Session management
Claude Code sessions
# Continue the last conversation
claude --continue
# Resume a specific session
claude --resume <session-name>
# Name the session during the conversation
/rename stripe-integration
Recommended aliases
# ~/.zshrc or ~/.bashrc
alias c='claude'
alias cc='claude --continue'
alias cr='claude --resume'
alias g='gemini'
alias cx='codex'
4. Git workflow
Auto-generate commit messages
"Analyze the changes, write an appropriate commit message, then commit"
Auto-generate draft PR
"Create a draft PR from the current branch's changes.
Make the title summarize the changes, and list the key changes in the body."
Use Git worktrees
# Work on multiple branches simultaneously
git worktree add ../myapp-feature-auth feature/auth
git worktree add ../myapp-hotfix hotfix/critical-bug
# Independent AI sessions per worktree
Tab 1: ~/myapp-feature-auth → new feature development
Tab 2: ~/myapp-hotfix → urgent bug fix
Tab 3: ~/myapp (main) → keep main branch
PR review workflow
1. "Run gh pr checkout 123 and summarize this PR's changes"
2. "Analyze changes in src/auth/middleware.ts. Check for security issues or performance problems"
3. "Is there a way to make this logic more efficient?"
4. "Apply the improvements you suggested and run tests"
5. Using MCP servers (Multi-Agent)
Key MCP servers
| MCP server | Function | Use case |
|---|
| Playwright | Control web browser | E2E tests |
| Supabase | Database queries | Direct DB access |
| Firecrawl | Web crawling | Data collection |
| Gemini-CLI | Large-scale analysis | 1M+ token analysis |
| Codex-CLI | Run commands | Build, deploy |
MCP usage examples
# Gemini: large-scale analysis
> ask-gemini "@src/ Analyze the structure of the entire codebase"
# Codex: run commands
> shell "docker-compose up -d"
> shell "npm test && npm run build"
MCP optimization
# Disable unused MCP servers
/mcp
# Recommended numbers
# - MCP servers: fewer than 10
# - Active tools: fewer than 80
6. Multi-Agent workflow patterns
Orchestration pattern
[Claude] Plan → [Gemini] Analysis/research → [Claude] Write code → [Codex] Run/test → [Claude] Synthesize results
Practical example: API design + implementation + testing
1. [Claude] Design API spec using the skill
2. [Gemini] ask-gemini "@src/ Analyze existing API patterns" - large-scale codebase analysis
3. [Claude] Implement code based on the analysis
4. [Codex] shell "npm test && npm run build" - test and build
5. [Claude] Create final report
TDD workflow
"Work using TDD. First write a failing test,
then write code that makes the test pass."
# The AI:
# 1. Write a failing test
# 2. git commit -m "Add failing test for user auth"
# 3. Write minimal code to pass the test
# 4. Run tests → confirm they pass
# 5. git commit -m "Implement user auth to pass test"
7. Container workflow
Docker container setup
FROM ubuntu:22.04
RUN apt-get update && apt-get install -y \
curl git tmux vim nodejs npm python3 python3-pip
RUN curl -fsSL https://claude.ai/install.sh | sh
WORKDIR /workspace
CMD ["/bin/bash"]
Safe experimentation environment
# Build and run the container
docker build -t ai-sandbox .
docker run -it --rm \
-v $(pwd):/workspace \
-e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \
ai-sandbox
# Do experimental work inside the container
8. Troubleshooting
When context is overloaded
/context # Check usage
/clear # Reset context
# Or create HANDOFF.md and start a new session
Cancel a task
Esc Esc # Cancel the last task immediately
When performance degrades
# Check MCP/tool counts
/mcp
# Disable unnecessary MCP servers
# Reset context
Quick Reference Card
=== Essential commands ===
/clear reset context
/context check usage
/usage check tokens
/init generate project description file
!command run immediately
=== Shortcuts ===
Esc Esc cancel task
Ctrl+R search history
Shift+Tab×2 plan mode
Ctrl+B background
=== CLI flags ===
--continue continue conversation
--resume resume session
-p "prompt" headless mode
=== Multi-Agent ===
Claude plan/code generation
Gemini large-scale analysis
Codex run commands
=== Troubleshooting ===
Context overloaded → /clear
Cancel task → Esc Esc
Performance degradation → check /context
