agentleFS
NVIDIAKnown publisherNVIDIA/nemoclaw-community/examples/demos/field/blender-omniverse-dgx-station/skills/ovphysx-host-runtime-boundary/SKILL.md

ovphysx-host-runtime-boundary

Skill142 starsChanged 3 months agoApache-2.0View on GitHub

Run configured native OVPhysX demos and report host GIF/status receipts when Hermes is sandboxed but the OVRTX/OVPhysX runtime and visible Blender process are on the host.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 3 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use this skill together with the official ovphysx-drop-contact-acceptance skill for this guide's split deployment.

    line 11
  2. Distinctive:

    When a task asks for the configured or guide-owned native OVPhysX demo, select this skill before generic scene-authoring skills.

    line 14
  3. Very specific:

    Do not pass fixture, output_dir, body_prims, or the words from the user request as path overrides.

    line 32
  4. Distinctive:

    Use the guide-installed host helper instead of constructing Python programs, reading host files from the sandbox, or copying native results …

    line 58
  5. Very specific:

    Do not run this snippet with sandbox terminal, execute_code, or a temporary Python file.

    line 77
  6. Very specific:

    Call the helper's preflight action from the already-open Blender process.

    line 112
  7. Very specific:

    Treat the host preflight as blocked only when Blender reports that its installed runtime is absent, unhealthy, or incompatible.

    line 116
  8. Very specific:

    Use the helper actions for normal execution.

    line 122
  9. Very specific:

    Never use sandbox read_file, terminal, or write_file on /home/...

    line 130
  10. Distinctive:

    Do not substitute Blender rigid bodies, keyframes, or inferred motion for a failed native run.

    line 138

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

OVPhysX Host Runtime Boundary

Use this skill together with the official ovphysx-drop-contact-acceptance skill for this guide's split deployment.

When a task asks for the configured or guide-owned native OVPhysX demo, select this skill before generic scene-authoring skills. Use the installed helper and its defaults immediately; the currently open Blender scene is not the fixture and does not need to contain stairs or rigid bodies.

For that configured-demo request, make one Blender MCP call with exactly this action and no overrides:

import runpy
from pathlib import Path
helper = Path.home() / ".local/share/nemoclaw-blender/ovphysx_host_helper.py"
runpy.run_path(
    str(helper),
    init_globals={"OVPHYSX_REQUEST": {"action": "run_configured_demo"}},
)

Do not pass fixture, output_dir, body_prims, or the words from the user request as path overrides. The helper reads the configured absolute host paths and returns the final native simulation and GIF receipt in one bounded action.

Topology

The always-on Hermes SOUL contains this deployment boundary. This skill adds OVPhysX-specific handling, but the same rule applies to every Blender task: use Blender MCP for Blender operations and host file access; sandbox terminal and file tools cannot see the Blender host filesystem.

  • Hermes and the installed public OV skills run inside the OpenShell sandbox.
  • Visible Blender, the installed add-on, ovphysx_grpc_server, and the ovphysx_bridge_client native extension run on the host.
  • Blender MCP is the approved control path from Hermes to host Blender.

The standard workflow does not copy the OV checkout or native runtime into the sandbox. Source-analysis tasks may add a sandbox-local checkout separately.

The native OVPhysX binary and extension are not expected under /sandbox. Their absence there does not prove that OVPhysX is unavailable. Do not run the standalone native probe directly with sandbox Python and do not ask the user to copy native runtime libraries into the sandbox.

Installed helper

Use the guide-installed host helper instead of constructing Python programs, reading host files from the sandbox, or copying native results across the boundary. The helper is additive to the OV add-on and is installed at:

~/.local/share/nemoclaw-blender/ovphysx_host_helper.py

Invoke it only through the Blender MCP tool named mcp_blender_execute_blender_code. If that tool is not visible yet, use tool search for blender execute code, then call the discovered Blender MCP tool. Pass the following snippet as that tool's code argument, changing only the request dictionary:

import runpy
from pathlib import Path
helper = Path.home() / ".local/share/nemoclaw-blender/ovphysx_host_helper.py"
runpy.run_path(str(helper), init_globals={"OVPHYSX_REQUEST": {"action": "preflight"}})

Do not run this snippet with sandbox terminal, execute_code, or a temporary Python file. Those tools execute inside /sandbox and cannot see the installed host helper. A successful helper action always appears as a mcp_blender_execute_blender_code tool call.

Supported actions are:

  • run_configured_demo: run the complete configured demo and return its final compact native simulation and GIF receipt. Prefer this for the guide demo.
  • preflight: inspect the installed host add-on and runtime.
  • prepare: run the configured fixture preparation script on the host.
  • preview: import the configured USD into visible Blender and render its starting state.
  • simulate: run native OVPhysX and write sampled authoritative poses.
  • replay: render those native samples in visible Blender and create a GIF.
  • status: return compact simulation and replay receipts.

For a complete request, call the needed actions in that order. Each action returns a compact JSON receipt. Do not fetch or print the complete pose timeline or native diagnostics through MCP.

The helper owns a dedicated Blender scene for each active fixture. Repeated preview and replay calls for the same fixture reuse that scene and restore its imported transforms before applying new samples. Do not clear the scene or re-import the fixture manually between helper actions.

The installed configuration supplies demo defaults. For another task, put only the necessary overrides in the request: fixture, body_prims, output_dir, steps, sample_every, fps, device, optional body_map, and optional render settings. A custom fixture must be a host-visible USD file. Scene authoring and choosing valid body prims remain governed by the upstream OVPhysX skills; do not guess them.

Preflight

Call the helper's preflight action from the already-open Blender process. It reports installed add-on diagnostics and resolved host paths without returning the complete runtime manifest.

Treat the host preflight as blocked only when Blender reports that its installed runtime is absent, unhealthy, or incompatible. A missing sandbox-local binary or .so is not a blocker in this topology.

Execution

Use the helper actions for normal execution. simulate runs a generic, guide-owned timeline adapter against the upstream add-on API and installed runtime without changing either one. It samples complete body states at bounded intervals and writes pose-timeline.json, ovphysx-report.json, and a compact status.json under the host output directory. replay accepts only a timeline labeled native-ovphysx-readback, renders each authoritative sample, and labels the output blender-replay.

Never use sandbox read_file, terminal, or write_file on /home/... paths. Never ask Blender to open a /sandbox/... path. If the installed helper is missing, report that setup blocker instead of recreating it ad hoc.

For Blender Python properties, enum values, operators, and animation APIs, use the blender-python-api-verification skill and inspect the running API through Blender MCP before mutation.

Do not substitute Blender rigid bodies, keyframes, or inferred motion for a failed native run. Render or encode a GIF only from authoritative OVPhysX pose readback, and label the rendering path accurately.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.