Kanner’s Agent Skills

A curated collection of Claude Agent Skills designed to enhance AI workflows with specialized capabilities.

English ｜ 中文

🎯 Available Skills

🚀 Prompt Optimizer

Transforms user-provided prompts into high-quality, clear, and effective instructions optimized for AI models.

Key Features:

• Systematic prompt analysis and optimization
• Applies comprehensive prompt engineering best practices
• Enhances clarity, specificity, and structure
• Improves AI model understanding and execution

When to Use: When you need to refine vague, unclear, or poorly structured prompts into effective instructions.

Location: prompt-optimizer/

---

🌐 Jta (JSON Translation Agent)

AI-powered JSON internationalization file translator with agentic reflection mechanism for high-quality multilingual content.

Source: This skill is based on the open-source Jta project
Requirements: Requires the jta CLI tool to be installed (the skill can auto-install it)

Key Features:

• Agentic Translation: AI translates, evaluates, and improves its own work (3x API calls per batch)
• Smart Terminology: Automatically detects and maintains consistent terms
• Format Protection: Preserves {variables}, {{placeholders}}, HTML tags, URLs, Markdown
• Incremental Mode: Only translates new/changed content (saves 80-90% API cost)
• 27 Languages: Including RTL languages (Arabic, Hebrew, Persian, Urdu)

When to Use: When you need to translate JSON i18n/locale files, add new languages, or update existing translations.

Location: jta/

Additional Setup:

• The skill will automatically check and install the jta CLI tool if needed
• Requires an API key from OpenAI, Anthropic, or Google Gemini
• See the Jta documentation for detailed setup instructions
• For skill-specific documentation, visit Jta Agent Skills Guide

---

🤖 UI Automation

Delegate tasks that need a rendered interface observed, a browser driven, or a desktop application operated to the external Codex agent — then recover the evidence needed to act on the result.

Key Features:

• Dispatches self-contained tasks through codex exec with a schema-constrained verdict
• Artifact contract: screenshots, recordings, and produced files land in a per-run directory so they stay retrievable — the executor’s return channel carries text only
• Requires a saved screenshot for every failure, so results are verifiable rather than merely asserted
• Playbooks for post-deploy UI acceptance and for driving native GUI applications