Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/rtk-ai/vox/claude-mdgit clone --depth 1 https://github.com/rtk-ai/voxWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/rtk-ai/vox/claude-md)<a href="https://agentmods.dev/instructions/rtk-ai/vox/claude-md"><img src="https://agentmods.dev/badge/instructions/rtk-ai/vox/claude-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00656 | $0.00656 |
| Opus 5 | $0.00328 | $0.00328 |
| Sonnet 5 | $0.00131 | $0.00131 |
| Haiku 4.5 | $0.00066 | $0.00066 |
Grade A, and why
vox CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 59 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Voice feedback (vox)
This project uses vox for spoken feedback.
After completing a significant task (feature, bug fix, refactor), provide an oral summary:
vox -l fr "Short summary of what was done, in French"
(-l fr routes to a backend with proper French voices; -b say only exists on macOS.)
Keep summaries under 2 sentences. Use French. Do not speak for trivial operations (formatting, single-line fixes).
Runtime: Bun
Toujours utiliser bun au lieu de npm, npx, node, ou tsx.
Roadmap: Mode conversationnel (inspiré PersonaPlex/Moshi)
Objectif: transformer vox d'un outil TTS one-shot en un assistant vocal conversationnel intégré à Claude Code / Codex, avec Claude comme cerveau (bien supérieur au 7B de Moshi).
Architecture cible
vox_converse mode:
Micro → Whisper STT (local) → Claude API (streaming) → TTS streaming (qwen-native) → speakers
Boucle continue avec VAD pour détection fin de parole
Modes: push-to-talk | vad | full-duplex
Latence cible: ~800ms première syllabe
Phases d'implémentation
| Phase | Feature | Description |
|---|---|---|
| 1 | vox_hear |
STT local avec Whisper (candle-whisper ou whisper.cpp bindings), VAD pour détection fin de parole |
| 2 | Streaming TTS | Parler pendant la génération Claude — alimenter qwen-native phrase par phrase |
| 3 | vox_converse |
Boucle conversation complète: hear → Claude API stream → speak stream. Nouveau tool MCP |
| 4 | Personas | .voxpersona = voice clone + style + system prompt. vox persona create "nom" |
Références techniques (PersonaPlex)
- Codec Mimi (Kyutai, MIT): SEANet + Transformer + RVQ 8 codebooks, 12.5 fps, 24kHz
- sphn lib: Opus streaming en Python — référence pour le streaming audio
- Protocole WebSocket binaire: 0x01+Opus (audio), 0x02+UTF8 (texte) — inspiration pour vox_converse
- Voice conditioning: embeddings .pt pré-calculés — enrichir voice cloning existant
- VAD pattern: silence detection pour turn-taking naturel
- Repo: https://github.com/NVIDIA/personaplex | Model: nvidia/personaplex-7b-v1
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 59 lines · 656 tokens per session scan A 8411c14062c8
vox CLAUDE.md is an instructions file published in the GitHub repository rtk-ai/vox (158 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 656 tokens to every session, about $0.0033 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
DeepSpeed AGENTS.md
AGENTS.md instructions for deepspeedai/DeepSpeed, covering deepspeed project rules, commit & ci requirements, code change discipline, test discipline and tool caveats.
pipecat AGENTS.md
AGENTS.md instructions for pipecat-ai/pipecat, covering agents.md, project overview, common commands, setup development environment and install pre-commit hooks.
next-sanity AGENTS.md
AGENTS.md instructions for sanity-io/next-sanity, covering agent notes, next-sanity/live export conditions, runtime nuance, ai literacy framework (ailf) and cursor cloud specific instructions.
relay AGENTS.md
AGENTS.md instructions for AgentWorkforce/relay, covering git workflow rules, never push directly to main, correct workflow, ... do work .. and stop here - let user merge.
icm CLAUDE.md
Instructions for rtk-ai/icm, covering projet: icm (infinite context memory), objectif, stack technique, structure du workspace and modèles de données.
moss AGENTS.md
Instructions for usemoss/moss, covering agents.md, what this repo is, repository layout, integrations & cookbooks and framework cookbooks (examples/cookbook/).