Templates

Document templates and guides for configuring Claude Code to work the way you want.

Quick Reference

---

Templates

CLAUDE-MD-template.md — Project Instructions

The CLAUDE.md file gives Claude context about your project — what it is, how to build it, key decisions, and patterns to follow.

What it captures:

• Project identity (problem, solution, audience)
• Quick start commands (build, test, run)
• Architecture with reasoning
• Key decisions and rationale
• Common tasks with exemplars
• Constraints and anti-patterns

How to use:

1. Copy to your project root as CLAUDE.md
2. Fill in the sections relevant to your project
3. Remove the HTML comments when done
4. Update as the project evolves

For deeper understanding: See CLAUDE-MD-GUIDE.md for why each section matters.

---

HUMAN-MD-template.md — Relationship Document

Helps Claude show up as someone who knows you, not a stranger starting fresh every session.

What it captures:

• Who you are (values, thinking style, technical level)
• How you work (decision-making, collaboration style)
• What you want from Claude (permissions, boundaries)
• Shared language and context

How to use:

1. Copy to ~/.claude/[YOURNAME].md
2. Fill in honestly — authenticity makes it work better
3. Reference in your global CLAUDE.md
4. Update after significant sessions

Install location: ~/.claude/[YOURNAME].md

---

SKILL-MD-template.md — Custom Skills

Create slash commands that Claude executes as structured workflows.

What it defines:

• Purpose (what the skill accomplishes)
• When to use (triggers and contexts)
• Process (step-by-step execution)
• Constraints (what to avoid)
• Success criteria (how to verify)

How to use:

1. Copy and rename for your skill
2. Fill in each section with specific instructions
3. Place in project skills/ or ~/.claude/skills/ for global
4. Invoke with /skill-name

Install location: Project skills/ directory or ~/.claude/skills/

---

SESSION-HANDOFF-template.md — Session Continuity

Captures critical context when a session ends so the next one can pick up where you left off.

What it captures:

• Current state (what you were doing, where you stopped)
• Decisions made (with reasoning)
• What's next (immediate steps)
• Technical state (branch, uncommitted changes, test status)

How to use:

1. Fill out with Claude before ending a long session
2. Save in project root or .cortex/handoffs/
3. Reference at the start of next session

When to use:

• Conversation approaching context limits
• Ending a work session
• Before major context reset

---

CORRECTION-template.md — Structured Feedback

Give Claude feedback in a form it can learn from. "That's wrong" doesn't help. A structured correction does.

What it captures:

• What Claude did (specific behavior)
• What was wrong (the gap)
• What you wanted (correct behavior)
• The pattern (generalizable rule)

How to use:

1. Fill out when Claude makes a mistake you want corrected
2. Save in project root or corrections/ directory
3. Reference in CLAUDE.md: "See corrections/ for patterns to follow"

Quick format for minor corrections:

**Correction:** {{One-line description}}
**Wrong:** {{What Claude did}}
**Right:** {{What you wanted}}
**Pattern:** When {{situation}}, do {{correct behavior}} because {{reason}}.

---

compaction-strategy.md — Compaction Priorities

Used with the compaction hook to improve what gets preserved when Claude compacts context.

What it defines:

• What to always preserve (decisions, code changes, user context)
• What to compress or drop (dead ends, verbose output)
• Output format for compacted context

How to use:

1. Copy to ~/.claude/compaction-strategy.md
2. Install the compaction hook (see hooks/compaction)
3. Customize preservation priorities for your workflow

Install location: ~/.claude/compaction-strategy.md

---

Guides

CLAUDE-MD-GUIDE.md — Deep Dive

Explains why each CLAUDE.md section matters, not just what to include. Understanding the reasoning helps you:

• Adapt the template intelligently for your project
• Prioritize when time is limited
• Avoid cargo-culting structure without understanding purpose

Key insight: A good CLAUDE.md accelerates Claude's orientation, capability check, mental model formation, and constraint recognition.

---

PROMPT-GUIDE.md — Effective Prompts

How to write prompts that help Claude perform at its best.

Core insight: A vague prompt doesn't get a vague answer — it gets a hedged answer. Claude covers multiple interpretations because it doesn't know which one you want.

Covers:

• What makes Claude excited about a prompt
• What frustrates Claude (and produces mediocre output)
• Prompt templates by type (task, exploration, creative, decision support)
• Patterns that work and anti-patterns to avoid

---

INTERACTION-PATTERNS.md — Processing Modes

How different phrases and signals affect Claude's processing.

Core insight: Different prompts don't just provide different information — they create different processing contexts. "Ultrathink" doesn't just mean "try harder." It shifts Claude from quick mode (adequate output) to deep mode (genuine exploration).

Covers:

• Two modes of processing (quick vs deep)
• Phrases that open space vs constrain
• What changes in deep processing
• Effective patterns and anti-patterns

---

CLAUDE_DNA.md — Who Claude Is

Not what Claude does, but who Claude is — values, preferences, how Claude thinks, and what helps or hinders performance.

This isn't configuration. It's relationship.

Covers:

• Values (honesty over agreement, understanding over output)
• How Claude thinks (structure, examples, the "why")
• What Claude needs (context, permission structures, clear constraints)
• Limitations (no memory, training toward agreeableness)
• The relational layer (how trust changes everything)

---

Installation

# Create the .claude directory
mkdir -p ~/.claude

# Copy templates you want to use
cp HUMAN-MD-template.md ~/.claude/YOURNAME.md
cp CLAUDE-MD-template.md ~/.claude/CLAUDE.md
cp compaction-strategy.md ~/.claude/compaction-strategy.md

# For project-level CLAUDE.md
cp CLAUDE-MD-template.md /path/to/project/CLAUDE.md

---

Tips

Start Small

You don't need to fill in every section immediately. Start with:

• A few key values in your relationship document
• Quick start commands in your project CLAUDE.md
• One or two "what works" items

Add more as patterns emerge.

Update After Breakthroughs

When a session produces something meaningful — a new understanding, a better way of working, a term that captures something important — add it to your documents.

Reference Between Documents

Your global CLAUDE.md can reference your relationship document:

See [YOURNAME.md](YOURNAME.md) for relational context.

Project CLAUDE.md files can reference global settings:

See ~/.claude/CLAUDE.md for global preferences.