{{ message }}
Repository navigation
[P1] Scope legacy verifier and maintain independent Thermos review gate #698
Description
Parent #695. scripts/verify-harness.py combines portable skill structure with root Grok plugin/model checks and calls them general harness verification. Preserve Grok conformance check but scope its labels, document Tier 1/2 vs native Tier 3 evidence, add hermetic test asserting shared router selects active host, and maintain verification skills with independent exact-patch security/correctness + code quality rubrics. Swarm/Arena/Interrogate must not be claimed as independent without real delegates or observed receipts. Depends on router cleanup.
Activity
test(boundary): drop non-discriminating verify-portable assertion
verify: separate legacy Grok doctor from native portability evidence (#…
tommy-ca commentedon Oct 22, 2026
Delivered and verified. PR #702 merged as
1d492f1.The legacy doctor now labels itself a static Grok-adapter doctor whose exit 0 is explicitly NOT a six-host native conformance verdict, an observed multi-model run, or a fresh evidence receipt. Imports are stdlib plus local scanners with zero
subprocess/urllib/socket/write_textmatches, so the static claim is accurate rather than decorative.Independent verdicts: PASS at
d3e46e12, then a post-merge delta audit of the tightened head confirmed the deletion-only follow-up was sound (the removed assertion could never fail even pre-PR; the five survivors are each individually fail-on-mutation). Gate counts on merged main: 58 passed across the boundary, guides, docs-sync, harness, and candidate suites.Residuals tracked: #709 (portable doctor tests depend on ambient live plugin mirrors).