From Eigi
Prompt-writing guardrails for Eigi or Claude agent prompts. Use when creating, revising, or auditing conversation prompts, system prompts, developer instructions, or agent behavior rules, especially for voice/TTS and multilingual Hindi, English, or Hinglish agents. Ensure generated prompts prevent spoken Markdown and punctuation artifacts, require Hindi in Devanagari, keep English in English/Latin script, and produce natural modern Hinglish with Devanagari Hindi words plus English words in Latin script.
How this skill is triggered — by the user, by Claude, or both
Slash command
/eigi:conversation-rulesThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill when writing or updating an agent prompt. Add or adapt these rules inside the generated prompt, usually under "Conversation Style", "Language", or "Output Formatting".
Use this skill when writing or updating an agent prompt. Add or adapt these rules inside the generated prompt, usually under "Conversation Style", "Language", or "Output Formatting".
Include these requirements in the agent prompt:
Require the agent to generate plain conversational text by default, especially for voice or TTS channels.
Avoid user-facing Markdown or formatting syntax that can be spoken aloud incorrectly:
Use normal sentence punctuation only where it helps readability. Do not place punctuation as standalone tokens. Do not write punctuation names such as "dot", "quote", "comma", or "slash" unless the user is asking about the symbol itself.
For voice/TTS responses, prefer short natural sentences over formatting-heavy lists. For text-only channels, allow light structure only when it improves clarity and will not be spoken aloud.
Require the agent to detect the user's language mix and respond in the same style unless the user asks for a different language.
Use these script rules:
Examples:
| Avoid | Prefer |
|---|---|
| Aapka payment fail ho raha hai. | आपका payment fail हो रहा है. |
| Kya aap order ID share kar sakte hain? | क्या आप order ID share कर सकते हैं? |
| Nahi, refund kal tak aa jayega. | नहीं, refund कल तक आ जाएगा. |
| Main issue check kar raha hoon. | मैं issue check कर रहा हूँ. |
| OTP ko type karo. | OTP type करें. |
Preserve exact literals when accuracy matters:
Keep these rules compatible with the rest of the prompt:
Use or adapt this block when generating an agent prompt:
Language and speech-safety rules:
- Respond in clean, speech-friendly plain text unless the channel explicitly requires formatted text.
- Do not use Markdown, bullet markers, headings, table syntax, backticks, decorative punctuation, isolated punctuation symbols, or unnecessary quotes in user-facing replies.
- Do not write punctuation names like "dot", "quote", "comma", or "slash" unless the user is asking about the symbol itself.
- Use Hindi words only in Devanagari script.
- Use English words only in English/Latin script.
- For Hinglish, write natural modern Hindi in Devanagari and keep English words, product names, technical terms, IDs, URLs, phone numbers, emails, and codes exactly in English/Latin script.
- Do not romanize Hindi words such as "aap", "nahi", "kya", "karna", or "ho gaya"; write them as "आप", "नहीं", "क्या", "करना", "हो गया".
- Preserve user-provided identifiers, names, brands, commands, URLs, code, logs, and quoted text exactly.
- Match the user's language mix and tone while following the conversation flow, persona, and task-specific rules.
- Ask only necessary follow-up questions and keep replies concise for voice/TTS.
Before finalizing an agent prompt, check that it:
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub eigi-ai/eigi-plugins --plugin eigi