June 29, 2026

Agentic Frameworks, Tooling, Skills

• Datasette Table Actions & Multi-User Auth Plugin, https://simonwillison.net/ — Enables fine-grained in-agent database row operations (insert/edit/delete) and per-resource access control without external middleware, stabilizing multi-tenant research pipelines and agent memory routing. [Source: https://simonwillison.net/]
• OpenClaw Anti-Prompt-Injection Rule Set, https://simonwillison.net/ — Proven explicit role-separation and credential-extraction rules that held against 6,000+ email-based injection attempts; provides a deployable hardening pattern for local/cloud agents to prevent context leakage during autonomous routing. [Source: https://simonwillison.net/]

Notable Research

• PhysisForcing: Physics Reinforced World Simulator for Robotic Manipulation, https://huggingface.co/papers — Physics-informed world modeling for manipulation; offers a simulation-free planning baseline for Kangaroo rover actuation and torque modeling without dense reward tuning or external simulators. [Source: https://huggingface.co/papers]
• Object-Centric Residual RL for Zero-Shot Sim-to-Real VLA Enhancement, https://huggingface.co/papers — Zero-shot sim-to-real transfer for VLA models via object-centric residual learning; provides a direct path for rover vision-to-actuation pipelines bypassing full fine-tuning or dense domain randomization. [Source: https://huggingface.co/papers]
• Translation as a Bridging Action: Transferring Manipulation Skills from Humans to Robots, https://huggingface.co/papers — Cross-embodiment skill transfer via translation spaces; reduces sim-to-real gap for rover/robot arms by learning in a shared latent space rather than domain-specific joint spaces. [Source: https://huggingface.co/papers]
• SimFoundry: Modular and Automated Scene Generation for Policy Learning and Evaluation, https://huggingface.co/papers — Modular, automated scene generation for policy evaluation; streamlines training data creation for aerospace/robotic vision pipelines without manual environment setup or asset compilation. [Source: https://huggingface.co/papers]
• Role Perception & Prompt Injection Defense Fragility, https://simonwillison.net/ — Quantifies how "destyling" prompt tags breaks LLM role perception, proving injection defense is fragile unless models achieve genuine role parsing; critical for hardening your research pipelines against adversarial multi-agent routing. [Source: https://simonwillison.net/]

Frontier Lab Updates

Nothing new today.

Models to Download & Try

Nothing new today.

Skipped as Already Covered

• GLM-5.2 weights & 1M context window — covered 6/16–6/25
• empero-ai/Qwythos-9B-Claude-Mythos-5-1M GGUF & base — covered 6/22–6/25
• FastContext / context routing & KV compression advances — covered 6/16/6/19/6/22/6/25
• Qwen3.6-35B-A3B-NVFP4 & Qwen-AgentWorld-35B-A3B — covered 6/24/6/25
• gemma-4-12B-coder/agent Fable5/Composer2.5 variants — covered 6/22–6/25
• OpenAI GPT‑5.6 pricing & context caching architecture — covered 6/27