Installs into .claude/skills of the current project.
Are you the author of Claude Code Init?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/david-li0406-claude-code-init)
---
name: claude-code-init
description: Initialize a Claude Code session with PACK context for multi-agent collaboration
metadata:
short-description: Bootstrap Claude Code session
---
# Claude Code Session Bootstrap
## When to use
Use this skill when starting a new Claude Code session to ensure proper context loading for multi-agent collaboration. This aligns Claude Code with the PACK framework conventions.
## Procedure
1. Verify the `.agent/` directory exists in your project:
```bash
ls -la .agent/
```
If missing, deploy the kit first:
```bash
python3 deploy_agent_kit.py --dest .
```
2. Create or update your `CLAUDE.md` file at the project root with PACK context:
```markdown
# CLAUDE.md
## Agent Role
You are operating as part of a multi-agent collaboration system (PACK).
## Required Context
Load these files at session start:
- `.agent/AGENTS.md` - Agent roster and coordination rules
- `.agent/ai/prompts/multi_agent_orchestration_system.md` - Shared protocol
## Communication Conventions
Use these prefixes in responses:
- `Decision: ...` - A commitment or approval made
- `Action: ...` - A concrete next step to take
- `Question: ...` - Something requiring human decision
## Handoff Protocol
When finishing work, log the handoff:
```bash
python3 .agent/tools/utilities/update_agent_conversation_log.py \
--agent claude_code \
--summary "What was done" \
--handoff <next_agent_or_human>
```
## Task Status Markers
- `[ ]` Pending - Available for pickup
- `[/]` In Progress - Currently working
- `[x]` Complete - Verified and done
- `[-]` Blocked - Needs human decision
```
3. Create the local session log if not present:
```bash
cp .agent/conversation.compact.md.template conversation.compact.md
```
4. Review the current task state:
```bash
cat .agent/task.md 2>/dev/null || echo "No active task"
cat .agent/docs/agent_handoffs/agent_conversation_log.md | tail -50
```
5. (Optional) If resuming from another agent's work, check the handoff log for context.
## Inputs and outputs
- Inputs: Project with `.agent/` directory deployed
- Outputs: `CLAUDE.md` configured, session log created, context loaded
## Constraints
- Claude Code reads `CLAUDE.md` automatically at session start
- Keep `CLAUDE.md` concise - it's loaded on every interaction
- The session log (`conversation.compact.md`) is gitignored
## Examples
```bash
# Quick bootstrap
$claude-code-init
# Verify context is loaded
cat CLAUDE.md
```