🧠 Claude Code Guide for Agentic Terminal Development

1. 🔧 Initial Project Setup

Run /init on project start

Creates a Claude.md containing:

Project summary
Component descriptions
Dev environment setup
Code style guide
Tooling notes
Commit Claude.md to repo

Ask Claude to write and commit this file. Use write to file + commit flows rather than inline pasting.

Enable file-based workflows
Press Tab: auto-completes filenames
Press Shift + Tab: toggles auto-accept for file changes

2. 🧭 Workflow Planning & Execution

Step-by-step Planning Approach

Use o1-pro / o3-mini-high or Grok for "pre-plan" brainstorming
Launch Claude Code in repo
Explain high-level plan with milestones
Ask Claude to break the plan into diffs
Guide Claude one diff at a time, reorienting when needed

Prompt Claude to Think

Use explicit thinking modes to encourage depth:

think
think hard
think deeper
ultrathink
Use Plan-Implement-Review Cycle

For large changes:

Ask for a detailed plan before code changes
Review and approve/refine
Execute changes and tests
Commit with contextual messages

3. ⚙️ Commands & Customizations

Use /compact regularly

Compresses state for long sessions; improves response performance

Define custom slash commands

Add .md files to:

~/.claude/commands/ (global)
.claude/commands/ (project)
Use /permissions to manage trusted tools

Avoid permission fatigue by explicitly allowlisting safe operations

4. 🧩 File Editing Best Practices

Always specify the file explicitly, e.g.:

  Claude, can you increase the input size in `user/page.tsx`?
  Claude, add error handling to `data_processing.py`.
  Claude, refactor the `calculate_totals` function in `invoice.py`.

Be explicit about file relationships
Write output to files for Claude to re-ingest

5. 🧠 Prompting & Reasoning Techniques

Few-shot prompting

Add 2–5 curated examples to Claude for higher output fidelity

Role definition

e.g., “You are a senior Rust compiler engineer” to shape tone & accuracy

XML tag prompting

Structure prompts as:

  <instructions>Write a data pipeline</instructions>
  <context>We use Dagster + S3</context>

6. 🧪 Testing & Quality Assurance

Test-Driven Development

Ask Claude to write tests before implementing features

Automated test running

Use Claude to repeatedly validate logic with local test runners

Review entire codebase

Periodically ask Claude to:

Audit for redundant code
Assess architectural consistency
Suggest cleanup/refactors

7. 🤖 Tool Integration via MCP

MCP Tools + Rules Strategy

Identify key workflows in your dev environment
Map available MCP tools (e.g., Puppeteer, Postgres, GitHub API)
Define routing rules for agent-tool coordination

MCP Headless Mode

Integrate Claude into CI/CD or scripts with non-interactive inputs

Multi-agent collaboration

Delegate tasks across Claude + other AIs in parallel

8. 🖼️ Visual/UX Workflows

Use:
Screenshots as visual guides
Claude for Figma-to-code translation
Feedback cycles comparing output to design

9. 📚 Ongoing Improvement Tips

Rotate LLMs (e.g., Grok, GPT-4o, Claude 3.5 Sonnet) for planning then hand off to Claude Code
Use a Markdown journal to document Claude’s performance patterns

gregsantos/claude-code-tips.md