Kelet vs SkillClaw

Kelet is a root-cause analysis agent for LLM applications that goes beyond trace visualization. Where most observability tools stop at showing you what happened, Kelet automatically reads your traces, cross-references failure patterns across thousands of sessions — thumbs-down ratings, abandoned conversations, LLM-judge flags — generates root cause hypotheses, and produces targeted prompt patches to address them. The workflow is: connect your traces (LangSmith, Langfuse, or direct API), let Kelet ingest your failure signals, and receive a prioritized list of failure clusters with explanations and draft prompt fixes. SOC 2 Type II certified, read-only access to traces — nothing is mutated. The indie team positions it as the missing "closing of the loop" in LLM observability: most teams can detect failures but have no systematic path from detection to fix. The HN thread surfaced a real pain point: teams know their chatbot is failing somewhere, but diagnosing which prompts, tools, or routing decisions are responsible requires manual trace archaeology. Kelet automates that archaeology and produces actionable output, not just dashboards.

SkillClaw is a research framework from Alibaba's AMAP-ML team that enables collective skill evolution for LLM agent systems deployed at scale. The core idea: instead of each user's agent interactions existing in isolation, SkillClaw aggregates anonymized skill-improvement signals across all users to continuously refine a shared library of reusable agent skills — without requiring centralized fine-tuning. The framework introduces a three-component architecture: a Skill Extractor that identifies and catalogs atomic capabilities from interactions, a Skill Evolver that proposes improvements based on aggregate feedback, and a Skill Selector that routes tasks to the best-available skill version per user context. Published on April 9 and hitting #1 on Hugging Face trending papers this week with 277 upvotes, the paper reports significant improvements over per-user baselines on complex multi-step agentic tasks. This matters especially for production agent deployments where cold-start problems are severe — a new user's agent immediately benefits from millions of prior interactions. It's a fundamentally different model of agent improvement than either fine-tuning (expensive, periodic) or RAG (retrieval-only, no learning).