tools/validation/README.md

Validation reporting — staging branch

Install the Claude Code plugin

The simready-report plugin ships as a Claude Code marketplace under validation/.claude-plugin/. From a checkout of this repo, run inside Claude Code:

/plugin marketplace add <path-to-repo>/validation
/plugin install simready-report@simready-playbook

That registers two slash commands: /simready-report and /simready-package. You still need a Python with the SimReady runtime deps and the foundations + sdk checkouts on disk — see To run the demo as-is below.

Read PROBLEMS.md before trusting any report

Reports produced by this prototype lean on a local patch for PROBLEMS.md P1: without it, every Robot-Body profile silently drops 4 of its 5 declared features at load time, because the foundation repo usd_profiles_codegen step doesn't read the JSON variant definitions next to each feature's markdown. A "passing" report against Robot-Body-Runnable would only have checked 1 feature out of 5.

The patch lives in plugins/simready-report/skills/simready-report/validate.py and is loud about itself — every run prints a [PATCH P1] line. With the patch active, Robot-Body-Runnable resolves to 5/5 features, 41 requirements instead of 1/8.

The dashboard's Caveats panel still surfaces the gap if anyone runs without the patch.

PROBLEMS.md P1 has the full root cause, the patch walkthrough, and the upstream fix path. The patch is meant to be temporary — once foundations codegen is fixed, delete _patch_register_json_variant_features() and its call site.

---

Status: demo / review staging. Not part of simready-explorer proper yet.

This directory is a verbatim copy of the prototype work currently living at loginowskid/simready-playbook, dropped here on the dev/dloginowski/validation-reporting branch so the SimReady team can review the artifact before we refactor it into the repo's existing library + skills patterns.

What this is

A Claude-Code-driven validation reporting pipeline for SimReady customer assets. On top of the existing omni.asset_validator engine and SimReady profile registry, it adds:

• HTML dashboard with summary cards, per-asset feature pass/fail, an asset filter, and a "Other failed requirements" panel for issues outside the loaded profile.
• External-dependencies provenance report — walks USD composition arcs per asset, classifies each layer as internal/external relative to the target, records actions taken to obtain each.
• Foundation doc rendering — renders the markdown source for every feature/requirement code referenced by the report into HTML inside the output.
• Auto-packaging convenience — for a raw asset bundle under assets_to_validate/<name>/, runs simready ingest usd per interface USD before validating against the packaged tree.
• Spec selector — --list-profiles enumerates registered profiles so the user can pick a SimReady profile interactively.
• Parallel asset validation — ProcessPoolExecutor workers, default auto.

Two slash commands ship together as a single Claude Code plugin:

• /simready-report — validate + dashboard + provenance + docs.
• /simready-package — standalone packaging (no validation).

Layout (mirrors the source repo)

validation/
├── bootstrap.ps1                    # Windows convenience installer
├── marketplace.json                 # Claude Code marketplace manifest
├── plugins/simready-report/
│   ├── plugin.json
│   └── skills/
│       ├── simready-report/         # HTML dashboard + validation
│       └── simready-package/        # standalone packaging
└── playbooks/
    └── foundations-deviations.md    # observed gaps between specs and asset_validator

Why it doesn't fit the repo's existing patterns yet

simready-explorer already ships source/libraries/validate/ with a clean public API (validate_asset, validate_asset_list, etc.) and three skills in the agentskills.io/specification format (validate-asset, validate-batch, validate-metadata). The work in this validation/ directory:

• Reinvents a thinner version of simready.validate.validate_asset_list — this is duplication that should be removed during integration.
• Uses a different skill format (Claude Code marketplace plugin) — should be rewritten as agentskills.io skills that call into the library.
• Is a standalone-script architecture (python validate.py …) — should be exposed via the existing repo.bat / library entry points.

Proposed integration after review

1. New skill validate-report under source/libraries/validate/src/validate/skills/validate-report/ — instructional doc that drives a new simready.validate.report module.
2. New module source/libraries/validate/src/validate/report/ containing the dashboard renderer, doc renderer, external-deps tracker, thumbnail mirrorer — extracted from report.py + external_deps.py and adapted to consume AssetValidationResult instead of the raw omni.asset_validator issue stream.
3. Optional new skill validate-package — thin wrapper around simready ingest usd (independent of validation).
4. Drop bootstrap.ps1 / marketplace.json / Claude-Code-plugin layout — not relevant here; the team uses repo.bat + _build/ flow.

To run the demo as-is

The scripts in this directory are self-contained and need a Python with the SimReady runtime deps installed plus the foundations + sdk checkouts on disk. Easiest path on Windows: & bootstrap.ps1 (creates a venv at %LOCALAPPDATA%\simready\ and persists SIMREADY_PYTHON / SIMREADY_FOUNDATIONS_PATH / SIMREADY_SDK_PATH env vars). On other systems or custom layouts, install the deps into your Python and set those env vars manually.

After that:

python plugins/simready-report/skills/simready-report/validate.py \
    <path-to-asset-bundle> --profile Robot-Body-Isaac --version 1.0.0

Output lands at <bundle>/.reports/<bundle-name>.<profile>/index.html.