Parent: #80171
PR: #80323
Confidence proof tracker: #80936
Staging repo: https://github.com/electricsheephq/openclaw-parity-harness

TLDR

The standalone Codex Desktop plugin wrapper exists and now exposes the latest confidence profiles. It still delegates to upstream OpenClaw QA-lab; it does not duplicate harness logic.

Latest plugin repo head:

electricsheephq/openclaw-parity-harness@2ce8e33
commit: Expose confidence proof plugin commands

Product impact: P4. This is an operator/reviewer convenience wrapper, not OpenClaw runtime behavior.

QA impact: P2/P1. It makes maintainer/local proof runs easier and reduces reliance on repo-specific agent memory.

What Exists

A standalone plugin repo was initialized at electricsheephq/openclaw-parity-harness with:

• .codex-plugin/plugin.json
• skills/run-openclaw-parity/SKILL.md
• skills/audit-openclaw-parity/SKILL.md
• scripts/run-openclaw-parity.mjs
• README.md and package self-test scripts

Supported Profiles

The wrapper supports:

• first-hour
• maintainer-gate
• tool-defaults
• tool-coverage
• soak-100
• jsonl-replay
• compare-harnesses
• confidence-100
• confidence-report
• confidence-self-test
• negative-controls alias
• fault-injection
• codex-native-live
• live-first-hour
• all-artifacts

confidence-100 is mock/offline-first. It only runs live/OAuth lanes when the user passes --include-live; otherwise the upstream confidence report should classify live lanes as environment-blocked.

Validation

node scripts/run-openclaw-parity.mjs --self-test
node scripts/run-openclaw-parity.mjs confidence-report --dry-run --openclaw-root /Volumes/LEXAR/repos/openclaw-1 --artifact-root /tmp/openclaw-artifacts
node scripts/run-openclaw-parity.mjs confidence-self-test --dry-run --openclaw-root /Volumes/LEXAR/repos/openclaw-1

All three commands passed locally after the latest update.

Boundary

The plugin is only an operator wrapper:

• It discovers a local OpenClaw checkout or installed openclaw binary.
• It invokes upstream openclaw qa ... commands.
• It collects artifacts for Codex to summarize.
• It does not fork scenario definitions, classifiers, runtime parity logic, token reports, or confidence scoring.

Follow-up

If maintainers want this under the OpenClaw org, transfer or recreate electricsheephq/openclaw-parity-harness as openclaw/openclaw-parity-harness after review.