Hear your agent suffer through your code

· coding ai · Source ↗

TLDR

  • Endless Toil is a plugin for Codex Desktop, Codex CLI, and Claude Code that plays escalating human groans in real time as a coding agent reads increasingly cursed code.

Key Takeaways

  • Works as a local plugin installed via Codex Desktop UI, Codex CLI marketplace commands, or Claude Code’s /plugin install endless-toil@endless-toil.
  • Audio playback uses afplay on macOS or paplay/aplay/ffplay on Linux; degrades gracefully to printed scan results when no audio player is available.
  • Plugin structure and marketplace layout follow the OpenAI Codex and Claude Code plugin specs, making it a working reference implementation for both ecosystems.
  • Sounds are tiered by severity: groan, wail, and abyss – testable independently via a bundled Python script before wiring into an agent thread.
  • Requires Python 3.10+ and must be explicitly invoked in a new thread; it does not auto-activate globally on install.

Hacker News Comment Review

  • Commenters unanimously want a demo video before trying it; absence of a visual walkthrough is the top friction point raised across multiple replies.
  • The strongest extension request is correlating audio intensity with wasted tokens or wrong assumptions, not just static code quality – pointing toward runtime observability rather than static analysis.
  • The author’s framing of this as an “emotional observability layer” for AI-assisted development got gentle skepticism; most commenters engaged with it as a joke tool, not an enterprise signal.

Notable Comments

  • @fredley: wants audio that fires when agent assumptions fail, volume scaled to tokens burned on the wrong path – a concrete runtime-observability use case.
  • @rob74: “I wish the agents could hear me when I have to suffer through their code” – flips the tool’s premise and gets at the real symmetry problem.

Original | Discuss on HN