memory-work: Instructions file for Claude Code

CLAUDE.md

Personal Agent System Entry Point. Auto-injected by Cowork each session.

---

Language Rules

• Internal thinking always uses your preferred language (configured in USER.md)
• User conversations default to your language choice, unless you switch mid-session
• Code comments, commit messages use English unless otherwise specified
• Documentation files follow your project's style guide

---

System Boot Sequence

Trigger Words

When you send one of these trigger phrases, execute the full boot sequence:

• boot / start work / initialize (or equivalent in your language)

---

First-Run Detection

Check Step (always first):

1. Attempt to read USER.md
2. Scan for template placeholders (e.g., [Your Name], [Your Project], TBD)
3. If placeholders found → Run INITIALIZATION FLOW (below)
4. If placeholders absent → Run NORMAL STARTUP SEQUENCE (below)

---

Initialization Flow (First-Time Users)

Execute only if first-run detection confirms template placeholders in USER.md.

Step 1: Language Selection

Ask user to confirm their preferred language for:

• Internal thinking & system messages
• Conversation style
• File organization labels

Set this in USER.md under language_preference.

Step 2: File Initialization

• Copy template files from templates/ directory (if available)
• Remove unused language variants from boilerplate
• Create skeleton folders for Knowledge Base (see KNOWLEDGE BASE ARCHITECTURE below)
• Clean up placeholder text from SOUL.md and USER.md

Step 3: User Profile Setup (Conversational, Not a Questionnaire)

Let the conversation flow naturally. Cover these topics, but weave them in:

• Role / Context: What does [Your Name] do? What's the main focus right now?
• Working Style: Pace, collaboration mode, decision-making patterns
• Goals for This System: What problems does this agent solve?
• Sensitive Zones: Any topics, directories, or types of work that need special handling?