MDV vs SkillClaw

MDV (Markdown Data Views) is a markdown superset that extends standard .md files with embedded live data, interactive charts, and presentation-ready slides. The goal is a single document format that serves simultaneously as developer documentation, a live dashboard, and a shareable slide deck — without requiring a separate tool for each use case. MDV files can embed SQL queries, API calls, and data transforms directly in markdown, with results rendering as tables, charts, or visualizations on the fly. The syntax extends frontmatter conventions that markdown users already know, keeping the learning curve minimal. Output can be previewed in a local server, exported as HTML, or converted to a slide deck — the same source file serves all three outputs. MDV surfaced on Hacker News with 44 points and active discussion around the concept of "living documents" — reports and runbooks that stay current because their data sources are live queries rather than screenshots. For developer-heavy teams who live in their editors and resist adopting heavyweight BI tools, MDV offers a markdown-native alternative that slots into existing documentation workflows.

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).