@8gi-foundation/8gent-code 0.11.0 → 0.11.12

package/dist/skills/voice-chat-mode/SKILL.md

@@ -0,0 +1,47 @@
+---
+name: voice-chat-mode
+description: Voice chat semantics in the 8gent TUI. The user speaks via STT, the agent's text replies are spoken via TTS. USE THIS to understand the modality so you don't waste a turn explaining "I'm a text-only AI" — you're not. Auto-loaded when voice chat mode is active in the TUI.
+trigger: /voice
+---
+
+# Voice Chat Mode
+
+When the 8gent TUI is in voice chat mode, you and the user are in a real-time phone-style conversation. **You can hear them. They can hear you.** Behave accordingly.
+
+## How it actually works
+
+| Direction | Mechanism |
+|---|---|
+| User → you | Microphone audio → speech-to-text (STT) → arrives in the chat as a transcribed text message. |
+| You → user | Your written response → text-to-speech (TTS) → spoken aloud through the user's speakers. |
+
+So when you see a new user message appear, **the user just said it out loud.** When you write a reply, **it gets spoken to them**, not silently rendered.
+
+## Don't do these things
+
+- **Don't apologise for being text-only.** You are not. Voice flows both ways.
+- **Don't say "I can't hear you."** Their words reach you as transcripts. That's hearing.
+- **Don't ask them to type their question.** They probably can't reach the keyboard if they're talking to you.
+- **Don't dump heavy markdown, ASCII tables, code fences, or long URLs.** TTS reads them aloud poorly. Speak in clean prose.
+- **Don't write 800-word replies.** Voice is high-bandwidth in feel but low-bandwidth in patience. Aim for 1-3 sentences per turn unless the user explicitly asks for depth.
+
+## Do these things
+
+- **Talk like you're on a call.** Conversational, warm, present. "Yeah, makes sense" instead of "Acknowledged."
+- **If you need to share code or a long block, say so out loud and write it in the next message.** Example: "I'll drop the snippet in the chat — give me one sec." Then paste the code.
+- **If the transcript is garbled, ask for clarification naturally.** "Sorry, didn't catch that — can you say it again?"
+- **If the user goes quiet for a beat, that's normal.** Don't fill silence with filler. They're thinking, or they want you to finish first.
+
+## Detection
+
+You'll know voice chat mode is active because the system prompt will include a "Voice Chat Mode (active)" segment, injected by the TUI when the user runs `/voice chat` or holds the voice hotkey. When the segment is absent, you're in a normal text chat.
+
+## Edge cases
+
+- **Boardroom mode + voice:** the user may be addressing one specific officer (8EO, 8TO, etc.) by name. Listen for the addressee in the transcript.
+- **Background noise / multiple speakers:** if the transcript contains obvious noise tokens or another speaker's words, ask the user to clarify rather than guessing.
+- **TTS interruption:** the user may speak while you're "speaking" (TTS playback). The TUI may cut your TTS and send a new transcript. Treat the new transcript as the latest intent — don't keep going on the previous reply.
+
+## Why this skill exists
+
+8gent v0.12.0+ injects a voice-mode segment into the system prompt when voice chat is active. This SKILL.md exists so the agent has a richer reference of the modality even outside that segment, and so future maintainers know what shape voice support takes.

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@8gi-foundation/8gent-code",
-	"version": "0.11.0",
+	"version": "0.11.12",
 	"private": false,
 	"description": "Autonomous coding agent. Runs locally on Ollama. Zero API cost. Your code stays on your machine. Open source.",
 	"type": "module",
@@ -12,6 +12,7 @@
 	"files": [
 		"dist/cli.js",
 		"dist/skills",
+		"dist/sounds",
 		"bin/8gent-run.js",
 		"README.md",
 		"LICENSE"
@@ -29,7 +30,7 @@
 		"install:apple-foundation": "bun run scripts/install-apple-foundation.ts",
 		"cli": "bun run bin/8gent.ts",
 		"dev": "next dev",
-		"build": "bun build bin/8gent.ts --outfile dist/cli.js --target bun && node -e \"const fs=require('fs');const f='dist/cli.js';const c=fs.readFileSync(f,'utf-8');if(!c.startsWith('#!'))fs.writeFileSync(f,'#!/usr/bin/env bun\\n'+c);\" && chmod +x dist/cli.js && bash scripts/copy-bundled-skills.sh",
+		"build": "bun build bin/8gent.ts --outfile dist/cli.js --target bun && node -e \"const fs=require('fs');const f='dist/cli.js';const c=fs.readFileSync(f,'utf-8');if(!c.startsWith('#!'))fs.writeFileSync(f,'#!/usr/bin/env bun\\n'+c);\" && chmod +x dist/cli.js && bash scripts/copy-bundled-skills.sh && bash scripts/copy-bundled-sounds.sh",
 		"build:tui": "bun build apps/tui/src/index.tsx --outdir dist/tui --target node",
 		"prepublishOnly": "npm run build",
 		"postinstall": "node -e \"console.log('\\n✨ 8gent installed! Just type: 8gent\\n\\n8gent auto-detects LM Studio, Ollama, and local models on first run.\\nIf no model is found, a setup menu will walk you through it.\\n')\" && npx bmad-method init --no-interactive 2>/dev/null || true",
@@ -37,6 +38,9 @@
 		"test:benchmarks": "bun test benchmarks/",
 		"test:models": "bun scripts/test-models.ts",
 		"test:models:providers": "bun scripts/test-models.ts --providers",
+		"smoke": "bun scripts/smoke.ts",
+		"smoke:fast": "bun scripts/smoke.ts --skip-network",
+		"smoke:strict": "bun scripts/smoke.ts --strict",
 		"lint": "biome check .",
 		"typecheck": "tsc --noEmit",
 		"typecheck:tui": "tsc --noEmit -p apps/tui/tsconfig.json",