switching AI agents shouldn't mean pasting your entire backstory again Fulcra gives connected agents a shared context backend for your files, notes and agent work → a research agent leaves its findings → a build agent picks up the relevant context → you can change tools and keep the work you've already accumulated you can inspect updates, trace what changed and revoke an agent's access for a launch workflow, I'd keep the brief, research and decisions together, then let each authorized agent contribute back. the next agent gets a useful starting point, and you still have a record you can review that handoff is what interests me here. better models are useful; having to rebuild the brief for every tool gets old connect through Fulcra's MCP server or supported agent skills. the connections and permissions still need setting up 5 GB free to start ↓

this is f*cking insane... a guy reverse-engineered a smart home app and gave Claude Code control of his house now Claude drafts a floor plan from his device names, and a web page turns it into a 3D model with live device states open a real door → the door in the model opens. tap a camera → live video from that room here's how it works step by step: step 1 → his tool, tolpo, finds the Tapo lights, plugs, sensors and cameras on his local network step 2 → an MCP server connects them to Claude, so he can ask things like "dim the office lights" or "is any door open?" step 3 → Claude reads the device names and guesses which rooms they belong to. it understands room names in English, German and Spanish step 4 → it drafts a floor plan. he corrects the rooms and layout before saving it step 5 → a local web page draws the house in 3D and puts each device in its room step 6 → every device shows its live state, so he can see what's happening around the house step 7 → door sensors drive the model: open the physical door, and its virtual version opens too step 8 → tap a camera to watch it live. before video starts, tolpo checks the camera's privacy switch under the hood: → device control runs directly over his LAN, without TP-Link's cloud in the middle → it authenticates with his own Tapo account → he tested it on his own bulbs, plug, hub, sensors, doorbell and cameras → most of the project was built with Claude Code, including the protocol research prompt + tool to set up your own version in the first reply ↓

Full prompt: You are my build partner. Help me connect my TP-Link Tapo smart home devices to Claude using tolpo, then set up a live 3D view of my home. Work step by step, explain each step simply, and ask before installing anything. PROJECT github.com/signerless/tolpo Read the current documentation first. Use the existing project and check which features work with my device models. GOAL I can ask Claude to control my lights and check my sensors in plain English. A local web page shows a 3D model of my home, with each device in its room and its current state. When a door sensor reports open, the door opens in the model. Clicking an enabled camera shows its live video. HOW IT CONNECTS My computer and Tapo devices are on the same local network. tolpo authenticates using my own Tapo account and talks directly to supported devices over the LAN. Claude connects to tolpo through its MCP server. tolpo's local web server displays the 3D house in my browser. PIPELINE Step 1: check my operating system, installed software and Tapo device models. Tell me what's supported, what hasn't been tested and what I need to install. Step 2: help me configure credentials locally, then run a read-only scan. List the devices found and flag anything unreachable. Step 3: connect tolpo to Claude. Test reading device states with questions like "which lights are on?" and "is any door open?" Step 4: after I approve, test one light: switch it on, change its brightness, then restore its original state. Step 5: draft a floor plan from the device names. Infer room names in English, German or Spanish where supported. Ask me to correct room assignments, dimensions and positions before saving. Step 6: open the local 3D dashboard. Check that devices appear in the right rooms and that their displayed states match the physical devices. Step 7: help me enable only the cameras I choose. Check privacy mode before opening video, and verify playback in the dashboard. Step 8: test a door sensor and a light against the 3D view. Check how the dashboard reports an unreachable device. PRIVACY AND CONTROL RULES (hard) Keep passwords and tokens out of chat and the repository. Guide me through entering them locally. Start read-only. Ask before changing device states or settings. Respect camera privacy mode. Never disable it automatically. Explain which device data or camera images would be sent to Claude. Local device communication does not mean the AI model runs locally. Start with step 1: ask which operating system and Tapo devices I have, then tell me what software and accounts I need.

20 best fast-growing GitHub repos for building your own AI product team around Opus 5.5 in the last 2 weeks coordinate coding agents, give them memory, research, control browsers, build interfaces, animate scenes and generate audio 15 new repos + 5 weekly risers from the 11–28 oct scan ↓ 01 / AGENT SYSTEMS 1. Orca → manage a fleet of coding agents github.com/stablyai/orca 2. Ruflo → coordinate multi-agent workflows github.com/ruvnet/ruflo 3. Reason Network → connect teammates coding agents github.com/reason-machines/r… 4. durable-harness → resume agents after interruptions github.com/vercel-labs/durab… 5. git-ai → source control for parallel agents github.com/feder-cr/git-ai 02 / MEMORY + CONTEXT 6. Graphify → query your code and documents as a graph github.com/Graphify-Labs/gra… 7. ai-memory → carry project memory across coding agents github.com/akitaonrails/ai-m… 8. Graphicle → code graphs + memory in one Rust binary github.com/Graphicle-AI/grap… 9. Hy-Memory → shared recall across AI applications github.com/Tencent-Hunyuan/H… 10. TokenEconomizer → reuse experience to streamline agent runs github.com/microsoft/TokenEc… 03 / TOOLS + RESEARCH 11. earshot → local speech tools for coding agents github.com/Leonxlnx/earshot 12. axuse → accessibility-first browser control through MCP github.com/ctoldigital/axuse 13. imozuru → trace research back to primary sources github.com/MaryCache/imozuru 14. esp32-iphone-agent → AI control of a real iPhone github.com/onehumanbeing/esp… 15. agentplay → replay agent sessions as maps + timelines github.com/Entelligence-AI/a… 04 / BUILD + CREATE 16. HyperFrames → turn editable HTML scenes into video github.com/heygen-com/hyperf… 17. fx0 → describe a page, get an editable HTML prototype github.com/vercel-labs/fx0 18. blender-skills → Blender production guidance for coding agents github.com/EL4CTEO/blender-s… 19. audio-studio-mcp → generate local music + sound effects github.com/kanaworksai-hue/a… 20. rossark-motion-skill → create animated SVG for interfaces github.com/rossark0/rossark-… save this, then build the best harness for Opus 5.5 for any task ↓

Anthropic just released new agent guide: make every subagent earn its own context steal this 18-role blueprint for turning Opus 5.5 into an entire engineering team PLAN → TOOLS → RESULTS → REVISE one main loop keeps the goal, plan and evidence → procedures become skills → API actions become tools → reference material loads when needed keep separate workers for fresh parallel research, independent verification or a measured context/tool boundary merge roles that repeatedly share data. compare each fold on the same end-to-end tasks and inspect what breaks official migration kit: > git clone github.com/anthropics/single… copy its skill/ folder to .claude/skills/single-agent-loop/ then in Claude Code: Run the single-agent-loop skill 18 roles in the map. evidence decides which ones need their own loop save this, then read the full guide for Opus 5.5 Harness Engineering ↓