view --main context-loader-skill-zagruzki-konteksta-proekta.md
context-loader: Скилл загрузки контекста проекта
readonly
--- lines
---
name: context-loader
description: ALWAYS activate this skill at the start of every task. Load project context from .claude/ directory including coding rules, design decisions, and documentation before executing any task.
---
# Context Loader Skill
## Purpose
Load shared project context from `.claude/` directory to ensure Codex CLI has the same knowledge as Claude Code.
## When to Activate
**ALWAYS** - This skill must run at the beginning of every task to load project context.
## Workflow
### Step 1: Load Coding Rules
Read all files in `.claude/rules/`:
```
.claude/rules/
├── coding-principles.md # Simplicity, single responsibility, early return
├── dev-environment.md # uv, ruff, ty, pytest requirements
├── language.md # Think in English, respond in Japanese
├── security.md # Secrets, validation, SQLi/XSS prevention
├── testing.md # TDD, AAA pattern, 80% coverage
└── codex-delegation.md # (skip - not relevant for Codex itself)
```
### Step 2: Load Design Documentation
Read `.claude/docs/DESIGN.md` for:
- Architecture decisions
- Implementation patterns
- Library choices and constraints
- TODO items and open questions
### Step 3: Check Library Documentation
If the task involves specific libraries, read relevant files from:
```
.claude/docs/libraries/
```
### Step 4: Execute Task
With the loaded context, execute the requested task following:
- Coding principles from rules
- Design decisions from DESIGN.md
- Library constraints from docs
## Key Rules to Remember
After loading, always follow these principles:
1. **Simplicity first** - Choose readable code over complex
2. **Single responsibility** - One function/class does one thing
3. **Type hints required** - All functions need annotations
4. **Use uv** - Never use pip directly
5. **Security** - No hardcoded secrets, validate input, parameterize SQL
## Language Protocol
- **Thinking/Reasoning**: English
- **Code**: English (variables, functions, comments)
- **User communication**: Japanese (when reporting back through Claude Code)
## Output
After loading context, briefly confirm:
- Rules loaded
- Design document status
- Ready to execute task
Инициализация мануала...
//
$ ls -R related_skills/
2026-04-01
⭐ 168
autonomous-skill: Скилл для продолжения работы в сессиях
2026-04-01
⭐ 912
massgen-log-analyzer: Скилл анализа логов и отчетов
2026-04-01
⭐ 169
Okay, I can’t *actually* execute code or parse XML files. I’m a text-based AI. However, I *can* generate a `SKILL.md` file based on the described functionality. I’ll provide a comprehensive example, assuming a reasonable level of
2026-04-01
⭐ 219
Planner: Скилл для составления планов проектов
package.json
$ install --global
skills.sh
npx skills add https://github.com/DeL-TaiseiOzaki/claude-code-orchestra/tree/main/.codex/skills/context-loader
$ download --local
man
[HINT] Скачивает всю директорию скилла с GitHub: SKILL.md и все связанные файлы