Picks (5)
1. Medical AI skill libraries reach "thousand-item" scale — OpenClaw-Medical-Skills and scientific-agent-skills land the same day
Repo: https://github.com/FreedomIntelligence/OpenClaw-Medical-Skills ｜ ★ emerging
With 869 medical agent skills this is the largest open medical skill library to date, spanning biomedical and clinical workflows; a second library covering verified scientific skills appeared the same day. The competition has moved from "which model" to "who ships the skill pack plus MCP servers".
2. On-device clinical de-identification — OpenMed makes HIPAA compliance a local capability
Repo: https://github.com/maziyarpanahi/openmed
Clinical NER plus HIPAA PII de-identification, running 100% on-device (Apple MLX + Python), bundling 2,200+ medical models across 21 languages, with data never leaving the intranet.
3. Food and nutrition AI moves from "logging" to "auditing" — FoodTruth uses a vision model to break packaging claims
Repo: https://github.com/dasouvik122005/FoodTruth
An open-weight vision model reads both the front-of-pack claims and the back-of-pack ingredient and nutrition panels, then flags three classes of problem: hidden sugar (a "no added sugar" product containing maltodextrin with a GI of 110), "whey-protein dusting" claims, and similar wording tricks.
4. GitHub turns Copilot from "picking a model" into "orchestrating models" — HydraFusion Max
GitHub's 10-14 update to Copilot's HydraFusion orchestration added a Max tier, refreshed the pool (Claude Opus 5.5 / GPT-6 Astra / GPT-6.1 Sol / GPT-6 Luna), and allocates reasoning effort by stage — medium for drafting, high for review — with cascading fallbacks.
5. WiFi becomes a health sensor — RuView monitors vitals without a camera
Repo: https://github.com/ruvnet/RuView
Turns ordinary commercial WiFi signals into real-time spatial intelligence: breathing, heart rate and presence detection with zero video capture. It gained 197 stars on the day and represents a rare "privacy at no cost" route for health monitoring in bedrooms and bathrooms.
Trends
- The skill becomes the unit of distribution. In both coding and medicine, competition has shifted from models to "skill pack + MCP server"; the medical side already has an 869-item library.
- Local-first is the default in health. OpenMed, FoodYou and RuView all insist data stays on the device — regulatory pressure is reshaping technology choices.
- Diet AI is upgrading from a logging tool to a judgement tool. Deterministic nutrition rules (GI, DIAAS, regulatory citations) are being hard-coded into audit engines instead of left to the model to improvise.
- Open weights keep accelerating. Meta Muse Spark Open (816B) and StepFun Step 5 Preview (600B) both opened weights the same day, both aimed explicitly at long-horizon agent tasks.
- Open-source supply is outpacing consumption. Agent orchestration, skill libraries and context compression dominate Trending; selection and governance — security, audit, context cost — are becoming the new bottleneck.