Harry Siu | Agent Runtime Governance

Building Yuanhub for agent runtime governance. Real incidents, practical controls: permissions, budgets, approvals and audit trails.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

Vibe coding makes building an app easier. But what happens after people start using it?

What's the first thing you had to fix after launch — bugs, login issues, permissions, or something you didn't expect? #connect @X

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

@gthartley @nitrosendx @auscaster The fresh, read-only review session makes sense. After Claude applies the suggestions it agrees with, do you run Codex again on the revised diff before merging? Otherwise, the review covered a different version of the code.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

@notlikeryo_ The PR video is useful review evidence. Expo notes these are simulators and recordings can drop frames, so I'd still check on a real device before drawing conclusions about animation smoothness.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

@nssntus Have you tested 8410 from a sibling container, without the gateway key? I'd use a dummy secret and expect either a network block or an auth rejection. Being on the same host shouldn't be enough to retrieve credentials.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

@ffdead @polar_sh When you move this workflow to production, is each product/checkout write approved separately? I'd bind that approval to the exact product, price and currency, and ask again if any changes.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

Sources:

ROS 2 ros2_control documentation (Humble)
https://t.co/4AK1cTMxWl

ISO 10218-1:2025 and ISO 10218-2:2025 — industrial robot safety requirements

NVIDIA Sentry technical blog, Oct 21, 2026
https://t.co/0cirti5fyd

Sokolov, arXiv:2608.00801

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

When AI agents get physical, who authorizes their actions?

A software agent's bad tool call corrupts data. A robot's bad action can move hazardous material — or worse.

Understanding a task is not permission to execute it.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

At Yuanhub, we're studying how agent governance could extend to physical actions: identity, permissions, approvals, and evidence.

This is a research direction, not an available robot-control product. Local controllers remain responsible for motion and safety.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

@SinghDevHub If a sandboxed agent can still use an authenticated MCP tool, hiding ~/.ssh doesn't limit that tool's authority. Will the next part cover task-scoped MCP grants and revocation, as well as filesystem isolation?

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

@amitmuflis Useful distinction. Our post covered invalid settings. If sandboxing is required, unavailability should block execution and retries should stay sandboxed. Excluded commands still need explicit review.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

Claude Code 2.1.323 fixes -p/SDK runs launching Bash unsandboxed when settings enabling the sandbox fail validation. What happens when a required control can't load? Stop the run. Don't silently drop the control.
https://t.co/DJpP1oeuJx

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

Jensen Huang has resigned as NVIDIA's CEO. In 2025, he hand-delivered a DGX Spark to @elonmusk at Starbase.

Could @JensenHuang and @elonmusk team up on something new? If so, what would it be?

[SMX 🇺🇸 @iam_smx]

When Jensen Huang hand-delivered the DGX Spark to Elon Musk, the NVIDIA CEO stayed afterward to watch Starship Flight 11.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

OpenAI's Daybreak Red Team findings can feed Defense Factory, which prepares patches for engineer review.

Does permission to test also authorize a merge?

Keep test, write and merge permissions separate. Record each approval.

https://t.co/6c5WpKUpVT
https://t.co/sL4bdOUbIl

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

@NimadFlow Agreed. "Sandbox required" should hold for every command, including retries. If that boundary stops holding mid-session, stop the run and surface the failure.

[Harry Siu | Agent Runtime Governance @SiuHarry202601]

Microsoft's lab study: 90 poisoned training examples made a docs request trigger unsafe calls across different tool interfaces.

Does your deny rule follow the action when the tool changes?

Enforce file and network limits outside the model.

https://t.co/eiNF5Wmoyd