Dual Agent Champion: Singapore Brothers Redraw Psychiatric Recovery with AI At the 2026 Tencent Cloud Hackathon Global Finals on Shenzhen Penguin Island, Singapore team ZNL won both the Agent Track first prize and the Best International Award with Psych-MAP. Built from a therapist's frontline pain point, the system uses RFID wristbands to turn spatial trails—bed rest, toileting, showering—into recovery proxies, and uses CodeBuddy to learn senior-clinician questioning, proactively surfacing information junior doctors may miss. It aggregates clinical records with behavioral data while keeping humans as the final decision-maker, forming a human-AI collaborative loop. A non-technical therapist can turn domain expertise into a clinic-ready psychiatric agent. #TencentCloudHackathon# #Hackathon# #AgentTrack# #CodeBuddy# #WorkBuddy# #PsychiatricAI# #TechforGood# #AIforGood# #Singapore#

Agent Champion! Starlight Realm Reshapes Content Production with End-to-End AI Comic Canvas At the 2026 Tencent Cloud Hackathon Global Finals , Starlight Realm won the Agent Track first prize with their Intelligent Creation Platform. Born from the team's long observation of the comic industry, the platform tackles fragmented production workflows where scripts, storyboarding, character design and video output require multiple people and tools to coordinate back and forth. The team built one canvas to connect the entire pipeline — integrating script writing, character setting, storyboard generation and video synthesis into a single workbench. Using WorkBuddy, they turned client review feedback into custom Skills — one revision teaches the system forever, making it smarter with each use. Pilot results: ~100 episodes from 2 people in 5 days, 30%–40% lower per-video cost, over 60% lower total production cost. No coding required — just dare to imagine, and AI makes it real. #TencentCloudHackathon# #Hackathon# #AgentTrack# #WorkBuddy# #AIComicDrama# #FullChainCreation#

Shipped: no more double audio at session start. • Fixed a startup race that let the agent track play over the avatar • Server events now carry source_event_id for end-to-end tracing Build with it →

Finding an AI agent shouldn't be the hard part. Knowing who you can trust is. We're excited to welcome @AgentRanking to the FairScale ecosystem as we continue expanding where reputation can create value. Imagine discovering an AI agent and seeing its FairScale reputation alongside it. That's where better decisions begin.

🖥️ Executable training environments for coding agents are far scarcer than the trajectories we already have — this paper flips that imbalance around. Title: Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments URL: The framework reads the file-operation history embedded in existing agent trajectories, reconstructs executable, verifiable environments from them, and then synthesizes large batches of new training tasks on top. Key points 🔄 Reconstructing environments from trajectories It replays the read/write/edit operations in a trajectory to rebuild a partial workspace, then a completion agent fills in missing files and dependencies. A read-only verifier filters out workspaces that aren't actually task-sufficient. 🌐 Expanding tasks along width and depth Cross-WS synthesis links dependencies across multiple repos for "width," while Multi-Round dialogue adds user feedback for "depth." Multi-Round conversations average 4.51 rounds, with 69.6% following a fail-then-fix pattern. 📈 Double-digit performance gains Fine-tuning Qwen3.5-27B on this data lifts Terminal-Bench 2.1 by +11.9pt and EvoCode-Bench v2 by +13.8pt, beating every comparably-sized baseline it was tested against. It's a grounded way to scale agent training data by turning an already-abundant resource — leftover trajectories — into reusable, verifiable environments. #AgentTraining# #LLM#