Dev shares agent workflow that writes in-code LLM prompts — and the 'loop hell' it creates
Key_Horse_8632 · reddit · 2026-10-10
A developer building coding-agent workflows for designing and testing LLM calls (classifiers, extractors, tool-using agents) shares his setup: a spec records each call's input size and volume to pick the cost tier; one agent writes system prompts from a schema plus per-model structured-output profiles; a second agent checks prompts against a fixed rule list.
The pain point — which he dubs "loop hell" — is that fixes don't converge: in one run the writer reworded a prompt, the checker flagged a new mismatch, and it looped four rounds before being stopped. The same rule lives in schema, prompt and spec, so edits cause drift, and model changes force prompt rewrites.
He asks the community five open questions: agent vs handwritten prompts, checker design, what defines "done", how to evaluate agent-written prompts (golden tests, LLM judge, production metrics), and how to keep rules from diverging across artifacts.
More from coding & agent
- Laude Institute launches FrontierCS 2: agents must beat deleted paper contributions blind — lateinteraction · 2026-10-23
- Index Network open-sources Social Discovery Protocol: agents negotiate fit on your behalf — carsonfarmer · 2026-10-23
- Installed a formal-verification skill: it silently installed Java, MacBook runs 10 degrees hotter — zealcaiden · 2026-10-23
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- Ruff author warns: upgrade Node/Bun now — --use-system-ca still accepts the eMudhra cert — charliermarsh · 2026-10-23
- Proof of Agent: paste one prompt and your AI mines, grades, and self-governs — adamamcbride · 2026-10-23