Hottest Repos
addyosmani/agent-skills — 3,275 stars today (54.4k total). New to digest. Today's biggest mover. Addy Osmani's production-grade engineering skills for AI coding agents — 24 skills across 6 dev phases (Define → Plan → Build → Verify → Review → Ship), 7 slash commands, 4 specialist agent personas. Works with Claude Code, Cursor, Gemini CLI, Codex, OpenCode, Kiro, and more. This is the "agent skills as a curated framework" approach vs. the marketplace model. At 54k stars in what appears to be early days, this could become the de facto engineering skills standard. MIT.
apple/container — 2,419 stars today (32k total). Second day, accelerating hard. From 29.5k → 32k overnight — nearly doubled yesterday's gain (1,358 → 2,419/day). Apple's Docker alternative using lightweight Linux VMs on Apple Silicon. OCI-compatible. The developer community is voting with stars: this trajectory suggests 35k+ by weekend. Apple is serious about owning the dev toolchain.
phuryn/pm-skills — 1,944 stars today (16.1k total). Third day tracking: 13.3k → 14.8k → 16.1k. Expanded to 100+ agentic skills and plugins for product managers. The "skills as products" wave has officially jumped from engineering to product management — and it's accelerating. MIT.
msitarzewski/agency-agents — 1,434 stars today (111.4k total). Complete AI agency with specialized expert agents. At 111k stars and still pulling 1.4k/day, this is the "agentic teams" pattern at massive scale.
obra/superpowers — 1,323 stars today (224.7k total). The gravity well continues. Gaining 1.3k/day at 225k stars. More infrastructure than project at this point.
AI & ML
chopratejas/headroom — 13,062 stars this week (23k total). Still the biggest weekly mover. From 6k to 23k in two weeks — context compression's poster child. Dropped off daily trending but the weekly numbers are staggering. The "context engineering" infrastructure layer is validated and entrenched.
mvanhorn/last30days-skill — 890 stars today (39.7k total). Seventh day tracking. Decelerating. From 2.5k/day peak → 890/day. The parabolic phase is over, but 11.7k stars this week and nearly 40k total means this is now a permanent fixture. The skills-as-products category champion.
maziyarpanahi/openmed — 427 stars today (2.7k total). Third day, holding steady. Local-first healthcare AI with 1,000+ specialized models, HIPAA Safe Harbor PII de-identification, native Apple MLX. No patient data leaves the device. Consistent growth without viral spikes suggests real practitioner adoption. Apache-2.0.
hexo-ai/sia — 177 stars today (1.2k total). New to digest. Self-Improving AI framework — three-agent loop (Meta, Target, Feedback) that autonomously improves task-specific agents across generations. Claims 56.6% gain on LawBench, top ranking on MLE-Bench Hard. The "AI that improves AI" pattern is getting real implementations. MIT.
Anti-slop movement: ~89k combined — taste-skill (41.5k, +8k/week, shipped v2 rewrite), impeccable (~37.5k). Up from ~88k yesterday. taste-skill's v2 brings brief inference, design-system mapping, and strict pre-flight checks. The movement isn't just growing — it's maturing its tooling.
Developer Tools
NVIDIA/SkillSpector — 308 stars today (2.6k total). New to digest. The skill ecosystem's immune system. NVIDIA's security scanner for AI agent skills — 64 vulnerability patterns across 16 categories, two-stage analysis (static + LLM semantic), live CVE lookups. Their finding: 26.1% of skills contain vulnerabilities, 5.2% show likely malicious intent. As skills proliferate, this becomes essential infrastructure. Apache-2.0.
refactoringhq/tolaria — 604 stars today (15.3k total). Fifth day, remarkably consistent. Desktop markdown knowledge base app. 12.7k → 13.4k → 14.2k → 14.8k → 15.3k. ~600/day like clockwork. The "files-first knowledge management" category has its winner.
FareedKhan-dev/train-llm-from-scratch — 538 stars today (5.6k total). New to digest. Complete pipeline from data download to text generation, including full post-training alignment (SFT → Reward Model → PPO/DPO → GRPO). Trains on consumer GPUs. The "demystify LLM training" niche found its most complete entry.
anomalyco/opencode — 609 stars today (173k total). The open-source coding agent at 173k stars, still gaining 600/day.
Hidden Gems
activeloopai/hivemind — 402 stars today (1.1k total). Shared memory for AI agent teams. Auto-captures session traces, extracts patterns into reusable SKILL.md files, shares them across Claude Code/Codex/Cursor/Hermes. Claims 25% cheaper, 1.7x fewer tokens, 31% fewer turns on LoCoMo benchmark. YC-backed. The "agent memory as infrastructure" problem is getting serious contenders. Apache-2.0.
kenn-io/agentsview — 98 stars today (1.6k total). Local-first analytics dashboard for 20+ AI coding agents — tracks costs, tokens, and sessions across Claude Code, Codex, Cursor, Gemini CLI, and more. One Go binary, no accounts, SQLite-backed. The "how much am I spending on AI agents?" question finally has a proper answer. MIT.
LLMQuant/quant-mind — 336 stars today (1.2k total). Knowledge extraction framework for quantitative finance — ingests 500+ daily research papers, builds semantic knowledge graphs, enables natural language queries. Accepted at NeurIPS 2025 GenAI in Finance Workshop. The "AI for alpha research" niche is still emerging. MIT.
Notable Trends
The skills ecosystem is growing its own immune system. NVIDIA's SkillSpector finding that 26% of skills have vulnerabilities is the wake-up call. We now have: skills creation (superpowers, agent-skills), skills curation (pm-skills, last30days-skill), skills security scanning (SkillSpector), and shared skills memory (hivemind). This is an ecosystem maturing in real-time — from Wild West to infrastructure.
Apple is making a serious developer platform play. container going from 1,358 → 2,419 stars/day (accelerating, not decelerating) suggests this isn't just curiosity — developers are ready for a Docker alternative that's native to their hardware. Watch for Apple to connect this to Xcode Cloud and their broader dev ecosystem.