llm-cure: two markdown files that stop coding agents from regex-ing everything
Careless_Love_3213 · reddit · 2026-09-23
- Frustrated by agents mechanically reaching for regex where judgment (an LLM call) is needed, over-asserting tests on code they just wrote, and mocking everything else, the author distilled his learnings into two markdown files that steer agents toward better problem-solving.
- LLMBOOST covers the loop's broad habits: inspect the real situation, make uncertainty explicit, keep solutions understandable, verify against evidence outside the model's own confidence.
- LLMCURE adds four diagnostics: name a risk and choose how to handle it; identify problem type before picking a tool; try small alternatives and let evidence decide; measure what's checkable and seek independent views for judgment calls.
- Grown from repeated agent runs in one codebase, packaged as MIT-licensed npm CLI llm-cure that installs these principles into supported coding agents, with an optional Agent Skill for Claude Code and Codex.
More from coding & agent
- Dev finds Claude Code smoother at reviewing PRs than at writing code — JasonBotterill · 2026-09-23
- Reddit dev gates agent links with a hand-tuned keyword score instead of the LLM — Most-Agent-7566 · 2026-09-23
- Rabbit launches OS3, a cloud agent that drives Windows, Mac and Linux remotely — emmanuelvivier · 2026-09-23
- Opus 5.5 builds a bike ride in the browser — every tree and sound generated in code — prasenx · 2026-09-23
- Bug Hunt Bench: GPT-6 Sol (max) matches GPT-5.6 medium but trails Opus 5.5 — PawelHuryn · 2026-09-23
- Claude Opus 5.5 adds time budget: let the model decide how long to work on a task — JeremyNguyenPhD · 2026-09-23