blurt AGENTS.md

A set of instructions for Blurt, an on-device voice-to-text app for Apple Silicon Macs. Voice recordings are transcribed locally and pasted where the cursor is.

In plain words
What is it for?
Use it when installing, running, testing, formatting, releasing, or changing the single-file Blurt application.
Why use it?
It gives coding agents the project's operating details, commands, architecture, and release process so they can work without guessing.

Instructions file for CodexOpenCode

Install

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.

agentmods
npx agentmods add instructions/satyaborg/blurt/agents-md
Clone the repo
git clone --depth 1 https://github.com/satyaborg/blurt

Made for: Codex, OpenCode.

Per session 729 This file is loaded in full into every session.
When invoked 729 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00729 $0.00729
Opus 5 $0.00365 $0.00365
Sonnet 5 $0.00146 $0.00146
Haiku 4.5 $0.00073 $0.00073

Measured 2d ago against content hash ea74abb61958, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

blurt AGENTS.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 2d 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.

AGENTS.md · 65 lines

How it starts

The opening of the file, as written. The whole thing — 65 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md

This file provides guidance to coding agents (Claude Code, Codex, Cursor etc.) when working with code in this repository.

What is Blurt

On-device voice-to-text for macOS Apple Silicon. Single-file Python app - press right ⌘ to start, press it again to stop, and text pastes at the cursor. Uses Qwen3-ASR through MLX locally, no cloud/API.

Commands

# Dev install
uv pip install -e ".[dev]"

# Run
blurt

# Tests
pytest
pytest tests/test_blurt.py::test_get_model_mode_default  # single test

# Lint
ruff check .
ruff format .

# Release (tags + pushes, triggers PyPI publish via CI)
./release.sh 0.5.0

Requires portaudio system dep (brew install portaudio).

Architecture

Single package: blurt/__init__.py contains everything. No submodules.

Flow: main() → starts pynput keyboard listener → successive shortcut presses toggle start_recording() and stop_recording() → stopped audio is concatenated and saved to ~/.blurt/audio/ → the selected Qwen3-ASR model transcribes it → pbcopy+osascript pastes it → the transcript is logged to ~/.blurt/blurts.jsonl.

Key design details:

  • The fast (Qwen3-ASR-0.6B-8bit) and accurate (Qwen3-ASR-1.7B-8bit) models are lazy-loaded and pre-warmed in a background thread at startup
  • Transcription paste works by saving/restoring clipboard (pbcopy/pbpaste) and simulating ⌘V via osascript
  • Silence is filtered using audio amplitude and RMS thresholds; prompt echo and repetition detection filter unsafe output before paste
  • Global mutable state (recording, audio_buffer, stream, etc.) protected by threading.Lock
  • Versioning via setuptools-scm (git tags), no hardcoded version
  • Sound feedback files bundled in blurt/sounds/*.mp3

CLI Subcommands

blurt (run), blurt add <phrase>, blurt rm <phrase>, blurt vocab (list words), blurt log (view transcripts), blurt upgrade (self-update via pipx).

Vocab

~/.blurt/vocab.txt - one word/phrase per line. Passed to Qwen as transcription context to bias recognition.

Read the full file on GitHub · 65 lines

Changes

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.

  1. 2d ago First seen · 65 lines · 729 tokens per session scan A ea74abb61958

Subscribe to this mod's changes

blurt AGENTS.md is an instructions file published in the GitHub repository satyaborg/blurt (5 stars, last pushed 2d ago), licensed MIT. It adds 729 tokens to every session, about $0.0036 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 2026-08-31.