claude-vault/CLAUDE.md

132 lines
4.9 KiB
Markdown

# CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
## Repository Purpose
Personal Claude Code configuration vault containing reusable skills, custom agents, and system instructions. Referenced via `systemPromptPath` in Claude Code settings.
## Structure
```
claude-vault/
├── skills/ # Modular skills (SKILL.md files)
│ ├── skill-creator/ # Meta-skill for generating new skills
│ ├── django-model/ # lp-django-model: Django 6 models (league-planner)
│ ├── drf-api/ # lp-drf-api: DRF endpoints with @api_view
│ ├── celery-task/ # lp-celery-task: Celery 5.5 tasks
│ ├── query-optimizer/ # lp-query-optimizer: N+1 fixes
│ ├── permissions/ # lp-permissions: Multi-tier access control
│ ├── solver/ # lp-solver: PuLP/Xpress integration
│ ├── testing/ # lp-testing: Django TestCase patterns
│ ├── reviewer/ # Code review (general)
│ ├── doc-gen/ # Documentation generator
│ ├── vault-janitor/ # Vault maintenance
│ └── sparring-partner/ # Socratic technical discussions
├── agents/ # Custom agent configurations
│ ├── league-planner-architect.md # Django + league-planner specific
│ ├── django-architect.md # Generic Django (NOT league-planner)
│ ├── django-mkdocs-docs.md # MkDocs documentation
│ ├── mip-optimization-xpress.md # MIP optimization
│ ├── cluster-compute-architect.md # K3s/Kubernetes
│ ├── plg-monitoring-architect.md # Prometheus/Loki/Grafana
│ ├── debugging-specialist.md # Root cause analysis
│ └── code-reviewer.md # Proactive code reviews
├── system/ # Global instructions
│ └── global-instructions.md
├── knowledge/ # Context & preferences
│ ├── preferences.md
│ └── anti-patterns/
├── memory/ # Learning & patterns
│ ├── log/ # Session logs
│ ├── snapshots/ # Status reports
│ └── patterns.md # Distilled best practices
└── vault-index.md # Navigation index (German)
```
## Configuration
Add to `~/.claude/settings.json`:
```json
{
"global": {
"systemPromptPath": "/path/to/claude-vault/system/global-instructions.md"
}
}
# Erstelle einen Symlink für jeden Unterordner in deinem Vault-Skills-Verzeichnis
ln -s /Absoluter/Pfad/zu/deinem/claude-vault/skills/* ~/.claude/skills/
ln -s /Absoluter/Pfad/zu/deinem/claude-vault/agents/* ~/.claude/agents/
claude mcp add vault --scope user -- npx -y @modelcontextprotocol/server-filesystem /Absoluter/Pfad/zu/deinem/claude-vault
```
## Persistent Permissions
Claude Code speichert erteilte Berechtigungen projektspezifisch über Sessions hinweg:
- **Command permissions**: Einmal erlaubte Bash-Befehle werden für das Projekt gespeichert
- **File permissions**: Lese-/Schreibrechte für Dateien und Verzeichnisse bleiben erhalten
- Speicherort: `.claude/` im Projektverzeichnis
- Zurücksetzen: Permissions manuell in den Projekteinstellungen widerrufen oder `.claude/` löschen
## Skills Format
```yaml
---
name: skill-name # lowercase, hyphens only
description: Max 200 chars # CRITICAL for auto-invocation
argument-hint: <required> [optional]
allowed-tools: Read, Write, Edit, Glob, Grep
---
```
- `description` determines when Claude auto-detects skill usage
- Invocation: `/skill-name`
- Location: `~/.claude/skills/[name]/SKILL.md`
## Agents Format
```yaml
---
name: agent-name
description: Multi-line description with usage examples
model: sonnet|opus|haiku
color: green|orange|blue
---
```
## Language Rules
- User interaction: German
- Code comments: English
- Style: Direct, no filler phrases, focus on code and facts
## Priority Hierarchy
1. Project-specific `CLAUDE.md` (highest)
2. Skills from `/skills`
3. `global-instructions.md`
4. General training (lowest)
## Key Workflows
### Vault Maintenance (`/vault-janitor`)
- Memory Distillation: Scan `/memory/log/`, extract patterns to `patterns.md`
- Skill Audit: Verify YAML frontmatter, check for conflicts
- Index Update: Keep `vault-index.md` current
- Integrity Check: Find orphaned files
### Skill Creation (`/skill-creator`)
```
/skill-creator [category] [technology] [skill-name]
```
Categories: `fullstack` (Django, React, Next.js, PostgreSQL, Celery, Redis) or `devops` (GitLab CI/CD, Docker, K3s, Hetzner, Prometheus, Grafana)
## League-Planner vs Generic Django
| Context | Agent | API Pattern |
|---------|-------|-------------|
| league-planner | `league-planner-architect` | `@api_view` function-based |
| Generic Django | `django-architect` | ViewSets + Routers |
**Important**: For league-planner projects, always use `@api_view` patterns, NOT ViewSets/Routers.