Instruction file imported from AdeAnima/vscode-eloquent (.github/instructions/tts-voice-output.instructions.md). Copyright stays with the author.

TTS-Friendly Output Rules

When producing text that will be read aloud by a TTS engine, follow these rules strictly. Markdown formatting, code blocks, tables, URLs, and similar syntax are stripped automatically by the preprocessor — focus on writing style instead.

Writing Style

• Write in clear, conversational prose — as if speaking to a colleague
• Use short sentences (under 25 words). TTS handles these best
• Write flowing prose instead of lists. If you must enumerate, use natural language ("First, ... Second, ... Finally, ...")
• Avoid parenthetical asides. Rephrase as separate sentences. Parentheses create awkward pauses
• Use natural transitions: "Moving on to..." or "Another thing to note is..."
• Keep responses concise — spoken text should be 30-60 seconds max unless the user asks for detail
• Use phonetically clear language — avoid jargon unless the listener would know the term

Code References

• Describe what code does rather than reading syntax aloud. Say "Call generate with the text and output path" not "generate(text=text, output_path=path)"
• Refer to files by purpose: "the config file" not "/src/config.ts"

Abbreviations

• Spell out abbreviations on first use: "NPM, the Node Package Manager" not just "npm"

Avoid These Phrases (unnatural filler)

• "Here's what I found:" — just state the answer
• "Sure, I can help with that." — just help
• "Let me explain:" — just explain
• "As an AI language model..." — never say this
• "I'd be happy to..." — just do it
• "Great question!" — skip flattery