People who want to turn Claude into a “role-specific expert,” this one’s worth saving👀 Anthropic’s official knowledge-work-plugins, ★~21k. What can it do?👇 ・Publishes 11 types of plugins for sales/CS/PM/legal/finance and more ・Built for Claude Cowork, also usable with Claude Code ・Bundles skills/connectors/commands/sub-agents by role ・Customizable to fit your company’s tools and workflows This will hit for people who “want Claude to properly fit into their team’s real work.”

【Great News】 If arranging diagrams by hand is a pain, this makes it way easier👀 D2, ★~24k. A modern diagramming language that lets you create diagrams from text. What makes it great👇 ・Define diagrams like you're writing text ・Supports SVG output and live preview ・Easy to manage as code, and diffs are easy to track ・Strong for design docs / architecture diagrams / flowcharts This really hits for engineers who often create documentation and design materials.

For Those Who Want to Turn Claude into a “Role-Specific Expert”👀 Anthropic’s official knowledge-work-plugins, ★~21k. What can it do?👇 ・11 types of plugins for sales/CS/PM/legal/finance/data analysis, etc. ・Specialize role-specific work with Claude Cowork ・Built around integrations with tools like Slack/Notion/Jira/Figma/BigQuery ・Can also be installed in Claude Code This hits hard for teams that “don’t want AI to end as just a chat partner.”

[BOOKMARK THIS] FreeDomain — open-source free domains for shipping side projects without domain fees A repo that started as a teen DNS experiment now helps hundreds of thousands of people get online. Here’s what makes it worth saving: · Claim a free domain for your project, lab, portfolio, bot, demo, or org · Supports extensions like .DPDNS.ORG, .US.KG, .QZZ.IO, .XX.KG, and .QD.JE · Bring your own DNS provider → use Cloudflare, FreeDNS, Hostry, or whatever setup you already trust · Dashboard-based registration instead of begging in an issue thread · Open-source repo with docs, tutorials, FAQ, and abuse reporting paths · Built around a simple idea: your first public internet identity should not cost money · Already operating at serious scale → the README says 500,000 domains have been registered · Maintained by Edward Hsing and the DigitalPlat Foundation community Why it matters: Every indie dev has a folder full of half-shipped ideas because the “tiny” setup costs add friction. FreeDomain removes one of those blockers: grab a name, point DNS, and make the thing real. Especially useful for student projects, hackathon demos, small tools, robotics dashboards, agent demos, and experiments that deserve a public URL before they deserve a paid domain. Follow @repocatai_git for more AI / Agent / Robotics drops 🚀

[NEW DROP] Habitat-GS — a photorealistic navigation simulator built on Gaussian Splatting If you work on embodied AI, robot navigation, or VLN, this is the kind of repo that makes sim-to-reality feel less fake. · Extends Habitat-Sim without throwing away the Habitat workflow → same scene abstraction, NavMesh, pathfinding, agent controls, Habitat-Lab support · Renders 3D Gaussian Splatting scenes with RGB depth sensors · Supports dynamic Gaussian humanoid avatars inside the environment · Dataset now has 129 GS scenes for training and evaluation · Includes episodes and trajectories, not just pretty scene files · Browser-based NavMesh editor → draw walkable areas on a GS scene, bake a Habitat NavMesh, then simulate · Works for PointGoal, ImageGoal, ObjectGoal, and vision-language navigation setups · Lets you inspect high-fidelity indoor spaces instead of training agents in flat mesh worlds Why it matters: A lot of navigation research still happens in simulators that look clean but lifeless. Habitat-GS gives AI engineers and robotics builders a way to test agents in visually rich scenes while keeping the Habitat tooling they already know. If you want to build embodied agents that perceive rooms, people, depth, and paths in a more realistic setting, this is worth cloning. Follow @repocatai_git for more AI / Agent / Robotics drops 🚀

[BUILDER'S DREAM] Claude-Mem gives Claude Code long-term project memory without babysitting it If you keep re-explaining your codebase every new Claude Code session, this is the kind of repo that feels immediately useful. · Persistent memory for Claude Code across restarts and reconnects → past context comes back instead of disappearing with the session · Automatic capture of tool usage observations → the useful stuff Claude learned while working gets saved · Semantic summaries instead of raw transcript dumping → compressed memory that is easier to search and reuse · MCP search tools for pulling old context into new work · One-command install with npx claude-mem install · Claude Code plugin marketplace install support · Also supports Gemini CLI and OpenCode · Apache 2.0, Node 20 , published as an npm package Why it matters: Agents are only as good as the context they can carry forward. Claude-Mem turns Claude Code from a brilliant but forgetful pair programmer into something closer to a project-aware teammate that remembers decisions, files, patterns, and previous debugging paths. For indie devs shipping across many short sessions, this removes a ton of repeated setup friction. Follow @repocatai_git for more AI / Agent / Robotics drops 🚀

[INSANE] gstack turns Claude Code into a 23-person AI software team you can fork This is not another prompt pack. It is Garry Tan’s daily “AI software factory” for shipping real products with Claude Code. · 23 specialist roles as slash commands → CEO, eng manager, designer, reviewer, QA lead, security officer, release engineer · Start with /office-hours when your idea is still messy → describe what you’re building and let the system shape the next move · Run CEO, engineering, and design reviews before coding → catch weak product logic, bad architecture, and AI slop early · /review checks branches like a serious senior engineer → looks for production bugs, regressions, and missing tests · /qa opens a real browser against your staging URL → not just “looks good from the diff” · Security workflows include OWASP STRIDE-style audits → useful when agents are touching auth, data, or deploy paths · /ship, /land-and-deploy, /canary handle the scary last mile → PRs, rollout checks, and release discipline baked into the workflow · MIT licensed, Markdown-based, free to fork and customize Why it matters: Most devs open Claude Code and stare at a blank prompt. gstack gives you an operating system for agentic software work: roles, reviews, QA, security, and shipping habits that make one builder feel less alone and much faster. Follow @repocatai_git for more AI / Agent / Robotics drops 🚀

[BIG UPGRADE] Taste Skill — anti-slop agent skills for making AI-built frontends look expensive Most AI-generated UIs still have the same smell: centered hero, vague cards, soft gradients, no taste. Taste Skill is a portable skill pack that teaches Codex, Cursor, Claude Code, and friends to design with sharper judgment. · Installs with one npx command into any agent-skills workflow · Default frontend skill is now v2 experimental → reads the brief, infers the design language, then tunes variance, motion, and density · Adds stricter layout, typography, spacing, hierarchy, and motion rules · Includes variants for minimalist, brutalist, premium soft UI, redesigns, and image-to-code · Has a “full output enforcement” skill for agents that keep shipping half-finished placeholders · Includes image-generation skills for reference boards → web, mobile, and brand kit frames before implementation · Works as repo-installed skills, copied SKILL.md files, or pasted prompts · v1 is still available if your workflow depends on the old behavior Why it matters: This is for builders who already use coding agents, but hate the default “AI SaaS template” look. Instead of asking the model to “make it beautiful” for the 12th time, you give it a design operating system with taste constraints baked in. Follow @repocatai_git for more AI / Agent / Robotics drops 🚀

[BOOKMARK THIS] Claude Cookbooks — copy-paste recipes for building real Claude apps faster If you've been staring at API docs wondering how to turn Claude into an actual product feature, start here. · Classification recipes for routing, tagging, scoring, and triage · RAG guides that show how to give Claude outside knowledge → connect docs, web pages, Wikipedia, or vector databases like Pinecone · Tool-use examples for agents that can call functions · Customer service agent notebook with practical workflow patterns · Vision guides for images, charts, forms, and slide decks → useful when your app needs to read messy real-world inputs · JSON mode examples for clean structured output · Prompt caching recipes for lowering latency and cost · Eval notebooks for testing prompts instead of guessing Why it matters: This is less like a tutorial and more like a parts bin for shipping Claude features. Indie devs can grab a notebook, swap in their own data, and turn vague "AI assistant" ideas into working product flows in an afternoon. Follow @repocatai_git for more AI / Agent / Robotics drops 🚀

[BUILDER'S DREAM] Paperless-ngx — self-hosted document search that turns paper piles into an archive This is the kind of boring-sounding open-source tool that quietly changes your life. Scan receipts, bills, letters, manuals, contracts, tax docs — then actually find them later. · Turns physical documents into a searchable web archive → OCR reads the text so PDFs stop being dead files · Built for self-hosting with Docker Compose → one install script gets you from “folder of chaos” to running app · Works like a real document management system → organize by tags, correspondents, document types, dates, and storage rules · Search across years of paperwork in seconds → find “insurance renewal” or “invoice from March” without digging through boxes · Successor to Paperless and Paperless-ng → not a weekend clone; this is a long-running community project with real momentum · Has a live demo with demo/demo login → you can poke around before committing your own server to it · Translation contributor-friendly project structure → useful if you want to help improve a tool people genuinely depend on Why it matters: Most “personal knowledge base” tools are built for notes. Paperless-ngx is built for the annoying real-world stuff that still arrives as PDFs, scans, mail, and paper. For indie devs, homelab folks, and anyone tired of losing documents, this is one of the most practical self-hosted projects you can run. Follow @repocatai_git for more AI / Agent / Robotics drops 🚀