A host-portable RuView agent harness minted via MetaHarness and hardened per ADR-182. Published as @ruvnet/ruview@0.1.0 (bare `ruview` blocked by npm's typosquat filter → scoped fallback). What it does: - 6 fail-closed `ruview.*` tools (onboard, claim_check, verify, node_monitor, calibrate, node_flash) exposed as CLI verbs + a dependency-free MCP stdio server. - The "prove everything" rule made executable: `ruview.claim_check` flags untagged accuracy claims and the retracted "100%" framing. - 5 host-neutral skills (onboard/provision-node/calibrate-room/ train-pose/verify) + bundled .claude/ config + provenance manifest. Validated: 17/17 unit tests, live MCP handshake, `ruview.verify` ran the real verify.py to VERDICT: PASS, clean `npx @ruvnet/ruview` from registry. Packs to 16.7 kB / 21 files; kernel+host are optionalDependencies so the operator tools install lightweight. README: documented as the portable, multi-host companion to the in-repo plugins/ruview/ Claude Code plugin (not a replacement).
1.5 KiB
name, description
| name | description |
|---|---|
| verify | Prove a RuView result is real — run the deterministic SHA-256 proof and the witness bundle (ADR-028), and lint any claim for MEASURED-vs-CLAIMED honesty. |
verify
The "prove everything" skill. Nothing ships as validated without this.
Deterministic proof (Trust Kill Switch)
ruview.verify runs archive/v1/data/proof/verify.py: it feeds a reference signal
through the production pipeline and hashes the output against
expected_features.sha256. Must print VERDICT: PASS. If numpy/scipy changed the
hash, regenerate with verify.py --generate-hash then re-verify.
Witness bundle (ADR-028)
For a release-grade attestation:
bash scripts/generate-witness-bundle.sh
cd dist/witness-bundle-ADR028-*/ && bash VERIFY.sh # must be 7/7 PASS
Contains the Rust test log, the proof + expected hash, firmware SHA-256 manifest, and crate versions — a recipient can re-verify with one command.
Claim honesty
Run ruview.claim_check {text} on any report, README section, PR body, or model card
before quoting accuracy. It flags:
- untagged accuracy numbers (must be MEASURED / CLAIMED / SYNTHETIC),
- MEASURED claims with no reproducer cited,
- the retracted "100%/perfect accuracy" framing.
Firmware-specific
A firmware fix is not "hardware-validated" without a captured boot log on real
silicon (e.g. the v0.8.1-esp32 rev-v0.2 validation: running headless so CSI captures (#1000) + CSI filter upgraded to MGMT+DATA + a no-false-detect mmwave
probe). Do not merge or release on a build-passes signal alone.