i18n-agent-skill

i18n-agent-skill 🌐

[English] | 简体中文

Industrial-grade frontend internationalization (i18n) lifecycle automation for AI coding assistants.

i18n-agent-skill provides full-lifecycle orchestration for internationalization within AI-native development workflows. By leveraging deterministic AST parsing and a "human-in-the-loop" staging mechanism, it minimizes hardcoded string leakage while ensuring predictable and verified translation quality.

---

🔄 How It Works (Continuous Iteration)

i18n-agent-skill is designed to be a continuous companion during your development lifecycle. As you build new features or refactor old ones, the cycle repeats—getting smarter with every iteration.

1. Identify: The AI performs a "handshake" with your project to understand its unique tone (e.g., Financial Professional). Done once at setup.
2. Extract & Audit: Whenever you add new pages or logic, the AI precision-scans the AST to capture new strings and find missing translations.
3. Refine & Commit: AI experts polish your draft in a staging area. Once you Commit, the system learns these choices and locks them as "Policy."
4. Sync: Verified high-quality translations are projected globally to all other languages instantly.
5. Iterate: Developing a new feature? Jump back to Step 2. Your i18n assets now evolve in lock-step with your code.

---

🚀 Installation

Option 1: Ask Your AI (Recommended)

Copy the following prompt and send it to your AI coding assistant (e.g., Cursor, Claude Code, Gemini CLI):

One-Step Setup (Recommended)

Setup this i18n skill for my current project and initialize it: https://github.com/Shirolin/i18n-agent-skill

AI should clone the repo, run the installer, then initialize the project with a valid entrypoint. init detects project locale assets first and then chooses sync/bootstrap/diagnostic behavior. Do not assume ./i18n exists until the installer has generated a project-level proxy.

Option 2: Manual Installation

# Clone to workspace (Recommended)
git clone --depth 1 https://github.com/Shirolin/i18n-agent-skill .agents/skills/i18n-agent-skill
cd .agents/skills/i18n-agent-skill

# Run Installer
./install.sh       # Linux / macOS
.\install.ps1      # Windows (PowerShell)

After installation, run the initialization command in your project root:

# If your platform maps slash commands:
/i18n-init

# If the installer generated a project-level proxy:
./i18n init          # Linux / macOS
.\i18n.ps1 init      # Windows (PowerShell)

# Direct Python entrypoint:
python -m i18n_agent_skill init

Choose the Right Entrypoint

Use the command form that actually exists in your environment:

• Slash command: /i18n-init, /i18n-status, etc. Only when your MCP/platform maps them.
• Project proxy: ./i18n or .\i18n.ps1 only after the installer has generated them in the target project root.
• Direct Python: python -m i18n_agent_skill <command> always reflects the underlying CLI.

init --auto is currently a compatibility flag for guided setup. It does not auto-save persona or skip confirmation steps.

Updating

cd <install_path>/i18n-agent-skill
git pull
./install.sh       # Linux / macOS
.\install.ps1      # Windows (PowerShell)

The skill automatically detects when a newer version is available and notifies you.

---

🛡️ Key Features

1. Deterministic AST Parsing

Unlike fragile RegEx-based extractors, our engine uses Tree-sitter AST to navigate code structure.

• Structural Accuracy: Correctness in complex JSX/TSX nesting and template literals.
• Zero-Noise Isolation: Automatically ignores comments and non-UI code blocks.
• Multi-Format Parsing: Robust support for JSON, YAML, and JS/TS object-literal locale files.

2. Privacy Shield (Secure by Design)

Built for enterprise security. Your source code and sensitive data stay in your local environment.

• Local Masking: Automatically masks extracted text and returned code context before AI interaction. Basic mode covers emails and API keys; strict mode also covers IPs, phone numbers, and ID numbers.
• Deterministic Hashing: Tracks changes via local hashes without ever uploading content.

3. State-Based Quality Evolution

Manages the translation lifecycle to prevent regressions and improve phrasing over time.

• State Machine: Tracks every key from DRAFT to REVIEWED and APPROVED.
• Glossary Memory: Saves confirmed terminology in the local glossary for later reuse.
• Typography Linter: Built-in rules for CJK-Western spacing and professional punctuation.

---

🌍 Language Support Matrix

Language Family	Extraction (AST)	Translation (AI)	Typography Linting	Status
English / Western	✅	✅	✅	Production
CJK (ZH, JA, KO)	✅	✅	✅	Production
European (Latin)	✅	✅	✅	Stable
RTL (Arabic, Hebrew)	✅	✅	⚠️ (Bypass)	Beta (Sync only)
Other (Hindi, Thai)	✅	✅	⚠️ (Bypass)	Beta (Sync only)

Note: Professional typography rules (e.g., CJK spacing) are currently optimized for language families marked as "✅".

---

📖 Core Command Set (AI & Human Reference)

Command	Capability	Detailed Functional Description
/i18n-init	Initialization	Detects project i18n state first, then returns sync, bootstrap, reconcile, unsupported, or hard_fail with concrete next steps.
/i18n-status	Health Check	Verifies dependencies, environment isolation, and current VCS (Git) state.
/i18n-scan	Extraction	Precision scan of source code to find hardcoded strings. Use --path for specific components or --vcs for current Git diff hunks.
/i18n-audit	Gap Analysis	Compares locale files against source code to find missing keys and report unextracted hardcoded-string candidates.
/i18n-sync	Smart Staging	Generates translation proposals from a flat {key: translation} payload. It rejects optimize task files and other wrapped artifacts.
/i18n-commit	Apply Changes	Formally writes approved proposals to physical locale files and updates quality snapshots.
/i18n-cleanup	Debt Control	Report-only dead-key detection. Review keys first; do not run ad-hoc deletion scripts directly on locale files.
/i18n-audit-quality	Expert Audit	Generates a quality report focusing on phrasing, variable safety, and typography.
/i18n-pivot-sync	Projection	Optimizes target languages based on a familiar reference language (e.g., zh-CN).
/i18n-optimize	Optimization Task Export	Exports an optimize task file containing targets and dynamic_glossary. This file is not a sync payload and must not be passed directly to sync.

---

🧭 Recommended Workflows

Already-Internationalized Project

1. Run i18n init.
2. Confirm mode=sync.
3. Run i18n status.
4. Run i18n audit --lang all.
5. During day-to-day development, prefer i18n scan --vcs to measure only the current change set.

New Project / First i18n Bootstrap

1. Run i18n init.
2. Confirm mode=bootstrap.
3. Create the base locale file, usually src/locales/en.json.
4. Add target languages after the base locale exists.
5. Run i18n status, then i18n audit --lang all.

Ambiguous Existing Project

1. Run i18n init.
2. If mode=reconcile, stop there.
3. Review the reported locale root candidates.
4. Confirm the intended locale root before running audit, sync, or cleanup.

Optional Persona Workflow

If you want persona-aware optimization, do it explicitly:

1. i18n distill-persona
2. Human review and confirmation
3. i18n save-persona --data ...

Cleanup Safety

• i18n cleanup is report-only.
• It does not generate deletion scripts.
• Delete locale keys manually in small batches.
• Run VCS diff, type-check, and tests after each batch.

Translation Optimization Workflow

1. Run i18n optimize --lang <lang>.
2. Review the returned summary and pending_confirmation fields first.
3. Treat temp/opt_<lang>.json as an optimization task file, not a locale patch.
4. Read only its targets entries and produce a flat {key: translation} JSON object.
5. Run i18n sync --lang <lang> --data <pure-json-or-file>.
6. Run i18n commit --proposal <lang>, then i18n audit --lang <lang> and type-check/tests.

dynamic_glossary is reference context only. It must never be written into locale files. temp/opt_<lang>.json is an intermediate artifact. Review its cleanup guidance before deleting it.

🤖 Integration Blueprint

This skill follows the Agent Skills Open Standard (ADK) and installs to the universal path:

Scope	Install Path	Who reads it
Workspace	.agents/skills/i18n-agent-skill/	Any ADK-compatible agent in this project
Global	~/.agents/skills/i18n-agent-skill/	Any ADK-compatible agent system-wide

For platforms that don't read the ADK path natively, run ./install.sh --link-platforms to create optional symlinks (e.g., ~/.gemini/skills/, ~/.claude/skills/).

---

📂 Project Structure

i18n-agent-skill/
├── i18n_agent_skill/   # Core Python logic package
├── scripts/            # Automation: installers, cleanup, and wrappers
├── references/         # Knowledge base: AST engine, Privacy Guard, Linting specs
├── assets/             # Templates: glossary, persona blueprints
├── tests/              # Full suite: Unit, integration, and resilience tests
├── SKILL.md            # Execution protocol (v4.0 Spec)
└── pyproject.toml      # Dependency and project index

---

🛠 Engineering Quality

This project maintains industrial-grade standards through automated verification tools:

# 1. Full Quality Audit (Ruff Format/Lint + Mypy Type Check)
python scripts/check.py

# 2. Automated Test Suite (58+ Unit & Integration Tests)
pytest

# 3. Protocol Compliance (Optional)
python .agents/skills/agent-skill-creator/scripts/validate.py .

---

🔒 Security & Privacy

We guarantee that zero source code is uploaded to third-party servers. All AST parsing, de-identification, and suggestion generation occur locally. AI agents only receive masked snippets required for translation assistance under your explicit permission.

---

💖 Support the Project

If you find i18n-agent-skill helpful, please consider:

• Giving the project a Star ⭐ to show your support.
• Afdian: https://ifdian.net/a/shirolin
• Ko-fi: https://ko-fi.com/shirolin

---

---

🔍 Deep Dive

• Product Scenarios: Detailed 5-phase lifecycle and authority hierarchy (L1-L3).
• AST Engine: How our Tree-sitter integration ensures pixel-perfect extraction.
• Privacy Guard: Specification for local data masking and security.

📄 License

Licensed under Apache-2.0.