KORA Doctor: open-source CLI flags LLM calls that should never have happened
NervousPengwin · reddit · 2026-10-07
The author argues agent observability answers what was called and what it cost, but not which calls should never have happened. His Apache-2.0 CLI KORA Doctor, inspired by the AUDR benchmark, scans agent traces for repeated inference, missed cache/reuse, deterministic work sent to an LLM, expensive models doing trivial tasks, and over-orchestration.
He's upfront about limits: without prompt bodies, metadata can't prove two calls are semantically identical, so outputs are candidates to inspect, not verdicts. A production practitioner also told him the biggest waste isn't repeated calls but re-fetching the same tool data across steps — waste that starts before the next model call.
More from coding & agent
- Codex runs GPT-6 with far less reasoning than the API — up to 12x lower on max effort — RexDouglass · 2026-10-07
- gpt-4.1-nano as RAG judge barely beats chance: AUROC 0.603 on RAGBench — Giulio Zeloni · 2026-10-07
- Ora scanned 107,797 sites: average agent-readiness score just 41/100 — EdenEmarco177 · 2026-10-07
- Cheap Codex carpool accounts are risky: your code and keys pass through strangers' servers — sujingshen · 2026-10-07
- Free 155-page Vibe Coding handbook ships 151 prompts, idea to production — Mahmoud_Zalt · 2026-10-07
- 10 RAG Projects That Take You From Basic Retrieval to Production-Grade AI Systems — _jaydeepkarale · 2026-10-07