review-gatelisted

# /review-gate ## Host dispatch contract - Codex: `$skill`, Codex collaboration roles, and GPT-5.6 tiers. - Claude: `/skill`, Agent/Skill tools, and Claude aliases. - A bare `/skill` in this shared source denotes the Claude form; Codex dispatches the same named skill as `$skill`. Cross-model 3-pass review of the current git diff. It tries the opposite CLI first; a bounded active-host fallback is explicitly degraded rather than misrepresented as independent cross-vendor review. > **Harness rule:** Codex tries Claude first; Claude tries Codex first. Only a > read-only availability failure permits one active-host retry. > **Reviewer context contract (v1.6.0):** every reviewer runs FRESH — a new > process/sub-agent fed the artifact (diff, and where the reviewer is agentic, a > read-only repo) plus the fixed pass prompt, NEVER the author's reasoning, > conversation history, or plan rationale. This is evidence-backed, not vibes: > fresh cross-context review F1 28.6% vs 23.8% when the reviewer is shown the > production context, vs 24.6% same-session self-review (arxiv 2603.12123 — the > context-aware reviewer is statistically indistinguishable from self-review). > Two failure modes, both forbidden: leaking author context into a reviewer > dispatch prompt (anchors the reviewer to the author's blind spots), and > starving the reviewer of the ARTIFACT (diff/spec/code) in the name of "low > context" — fresh means no reasoning trail, not less artifact. See `docs/promotion-protocol.