devrage tallies agent frustration across AI assistants

It’s a gimmick with a sharp edge: profanity is a crude signal, but it captures human friction with agent harnesses better than a lot of polished productivity metrics.

• –The local-only scan keeps the tool grounded in real conversations instead of self-reported sentiment
• –Typos, variants, and repeat collapsing make the detector more robust than a simple grep for expletives
• –Support for Claude Code, Codex, OpenCode, Amp, Cline/Roo, Zed, and Pi turns it into a cross-agent comparison tool
• –Rage reports could become a proxy for trust, latency, and how often an agent pushes developers into repair mode
• –The joke lands because it points at a real problem: some assistants are more “copilot,” others are more “stress test”