Release 9.0.0: remove the deprecated /flow-reviewer command

.agents/plans/12-reviewer-removal-release/README.md

@@ -0,0 +1,55 @@
+# Offline qualification for the 9.0.0 reviewer command removal
+
+The operator explicitly authorized a provider-free qualification path for this
+release on 2026-09-15, after being told the alternative was a funded live
+campaign, and after being told that a fourth consecutive offline release is the
+weakest claim yet on an exception written for individually reviewed features.
+This path covers only 9.0.0 and reviewer-command-removal-v1.
+
+## What this release does
+
+Removes `/flow-reviewer`, which 8.6.0 announced as deprecated, satisfying the
+cadence rule that a surface is announced in one release and removed no earlier
+than the next major. It also renames the optional TUI module id from
+`opencode-plugin-flow.reviewer-picker` to `opencode-plugin-flow.model-picker`,
+which is public module identity and therefore held for the major.
+
+The candidate deletes one palette command and changes one identifier string. No
+domain logic, persistence, validation, review gate or agent permission changes.
+Saved reviewer and planning preferences are read from the same agent options, so
+upgrading requires no reselection.
+
+## Frozen scope
+
+`scripts/feature-release.ts` pins commit f2d92b80f9caab422646235e3d3eb7e7e319c361,
+which holds the final package manifest, README pins and changelog. The frozen
+file list is the same one 8.5.0 and 8.6.0 froze: only the reviewed commit
+advances. Every enumerated picker and planning source, server configuration
+adapter, prompt surface, flow-plan and flow-run guide, package manifest and
+build/type surface must match that commit's Git blob and regular-file mode
+exactly.
+
+Nothing in the offline path constrains the major or minor, unlike the patch
+path. That is what makes this release mechanically possible; it is not evidence
+that it is well measured.
+
+## Baseline and assurance
+
+The baseline is 8.3.0, the last release with a live multi-provider campaign and
+canary. Offline releases never become baselines. The retained 8.6.0 seal could
+not be reused because the grader closure covers `package.json`, so the version
+bump alone invalidated it; the same original 8.3.0 campaign and canary were
+regraded offline, with no model calls.
+
+**Five releases now sit between this candidate and the last live measurement.**
+Nothing in this record narrows that gap, and each offline release widens it.
+Candidate assurance is reviewed frozen code, full deterministic checks, all
+retained gated replays, package checks, and the provider-free OpenCode smoke.
+None of these measure planning quality or coding outcomes.
+
+## Recommendation for the next release
+
+Fund a live campaign. A qualified release becomes a new baseline, which would
+reset `OFFLINE_FEATURE_BASELINE`, restore the patch path that has been
+unavailable since 8.4.0, and end a disclosure that has grown from one release
+behind to five.