amicus 4.6.2 → 4.6.3

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "amicus",
-  "version": "4.6.2",
+  "version": "4.6.3",
   "description": "Multi-model LLM Council + parallel AI window for Claude Code. Run structured council reviews across Gemini, GPT, DeepSeek and more — or fork a conversation to any model and fold the results back.",
   "author": {
     "name": "Christian Wagner"

package/CHANGELOG.md

@@ -5,6 +5,61 @@ All notable changes to Amicus are documented here. Format follows
 
 ## [Unreleased]
 
+## [4.6.3] - 2026-08-05
+
+### Added
+
+- **`fable` now carries an authored direct-Anthropic route** (`anthropic/claude-fable-5`,
+  verified live: Anthropic's `/v1/models` lists it and a direct leg serves). With an
+  Anthropic key present, `fable` routes direct-first like the other Anthropic aliases;
+  the `ANTHROPIC_MODELS` floor gains a matching row so keyless installs validate it.
+- **`doctor` gains a `session-metadata-tmp` check; `--fix` sweeps the orphans.** A kill
+  between an atomic write's tmp-file and rename leaves `.metadata.json.*.tmp` orphans in
+  per-session directories (the B09 class — ~30 write sites). Plain `doctor` reports them;
+  `--fix` removes orphans older than 60 s from the current project's sessions root and
+  announces the heal in the one voice (`Recovered: …`).
+- **`council save` announces when it shadows a built-in bench.** Saving a council named
+  `free`/`budget`/`frontier` previously printed no notice at all (the overwrite marker only
+  tracked user-config names); the save now reports `shadowsBuiltin` (`--json`) and prints
+  the shadow notice.
+
+### Fixed
+
+- **`models --check` no longer false-flags deliberate gateway-only routes.** A curated
+  alias whose direct form is derived (not authored) from its OpenRouter route is no
+  longer reported STALE — flat row, `GATEWAY STALE` row, candidates, and the
+  `fix: --add-alias` retarget suggestion all suppressed together — when the OpenRouter
+  route still serves. Deliberately gateway-only entries (`gpt-pro`) are annotated in
+  the curated data and never audited for a direct sibling. Kills the v4.6.2
+  release-gate false positive whose suggested "fix" was a silent tier downgrade.
+- **The Workspace seats panel's dead-seat rows are now role-aware and old-run
+  resilient.** A model that died as critic but succeeded as chair no longer has
+  its dead row hidden by the chair's cost row (only a live reviewing leg —
+  seat/critic/lens — suppresses, and a dead critic only by a live critic leg);
+  the dead critic's row names its role. Pre-v4.6 runs render their losses too:
+  `verdict.seatLoss.deadBenchSeats` feeds rows, and `verdict.json`'s
+  `degrades[]` backstops a `run.json` that lost its checkpoint. A stale
+  `get-run` reply from a run you navigated away from can no longer repaint the
+  run now open.
+- **A valueless `-o`/`--out` on `council verdict` now errors** (`BAD_ARGS`, flag named,
+  exit 1) instead of crashing mid-write (renameSync `TypeError`) and orphaning a
+  `true.tmp-<pid>` temp file. Behavior change, per the v4.6.3 R1 ruling.
+- **README corrections from the v4.6.3 accuracy review**: the Node.js floor is
+  22.12 (required since v3.0 — the README, install scripts, landing page, and
+  `amicus doctor`'s node check still said 18), and the optional-council-elements
+  list now matches the shipped skill (four opt-ins; the chair's verdict scale
+  has been standard, not opt-in). The doctor's `node` row now errors below
+  22.12 (it previously passed anything ≥ 18).
+
+### CI
+
+- **The MCP-Registry skip-check now verifies the version it trusts.** The
+  release workflow's idempotency pre-check previously skipped registry publish
+  on a bare HTTP 200; it now also requires the response body to both name the
+  exact version and report it as `active`, so preview-API schema churn — or a
+  stale/deprecated registry entry — can no longer produce a false skip. Every
+  new failure mode still routes toward publishing.
+
 ## [4.6.2] - 2026-08-05
 
 ### Added
@@ -91,6 +146,17 @@ All notable changes to Amicus are documented here. Format follows
   with replacement suggestions by `models --check` / `doctor`. This was the pin that made
   `models --check` exit 1 — the v4.6.2 release-gate risk.
 
+### CI
+
+- **The macOS/node-24 unit leg now runs jest with `--workerIdleMemoryLimit=1GB`.** That leg —
+  and only that leg — intermittently lost a worker to a native SIGSEGV ("A jest worker process
+  was terminated by another process"), which fails whichever suite occupied the worker with
+  zero assertion failures and a green rerun. Three confirmed hits (2026-07-31
+  run-cost-unknown, 2026-08-04 PR #100 update-notice, 2026-08-05 PR #105 — 6466 passed, the
+  dead worker alone failed 1 suite) tripped the standing third-occurrence rule. The limit
+  makes jest recycle an idle worker before the leak reaches segfault territory; the flag is
+  injected via a matrix `include`, so the other five legs still run a bare `npm test`.
+
 ## [4.6.1] - 2026-08-03
 
 ### Added

package/README.md

@@ -12,7 +12,7 @@ Hand Claude a plan, a design, a diff, an architecture decision, a manuscript —
 
 [![npm version](https://img.shields.io/npm/v/amicus?color=D97757&labelColor=1A1C29)](https://www.npmjs.com/package/amicus)
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue?labelColor=1A1C29)](./LICENSE)
-[![Node.js](https://img.shields.io/badge/node-%3E%3D18-brightgreen?labelColor=1A1C29)](https://nodejs.org)
+[![Node.js](https://img.shields.io/badge/node-%3E%3D22.12-brightgreen?labelColor=1A1C29)](https://nodejs.org)
 [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg?labelColor=1A1C29)](./CONTRIBUTING.md)
 
 **[Quick start ↓](#quick-start)** · [Commands](#commands) · [Documentation](#documentation) · [Troubleshooting](#troubleshooting)
@@ -99,12 +99,11 @@ flowchart LR
   findings-by-tier, cost — no chair prose). This is the default artifact handed to the user.
 - For an **editable source**, the accepted edits land in `<stem>-reviewed.<ext>` next to the original.
 
-**Optional council elements** (v2.2.0, all default off): five opt-in behaviors, offered once as a menu at launch — nothing turns on unless you name it, and the confirmation lists exactly what's on.
+**Optional council elements** (v2.2.0, all default off): four opt-in behaviors, offered once as a menu at launch — nothing turns on unless you name it, and the confirmation lists exactly what's on. **Chair verdict scale** *(standard since v2.2.0's follow-ups — no longer opt-in)*: the chair always closes with 3–5 hard questions and one parseable `VERDICT: Ship it | Fix these first | Fundamental rethink` line.
 
 - **Critic seat** — one reviewer swaps to a four-pass adversarial brief (adversarial pass, edge-case hunt, consistency check, executability test). Its findings enter the same anonymized bundle as everyone else's, so the bench disciplines the critic: manufactured negativity lands Disputed and dies in the tally.
 - **Expert lenses** — each reviewer takes a distinct expert perspective; you pick the panel domain (business, technical, customer, financial, or custom). Lens runs never feed the reliability ledger, and the report discloses the weakened cross-review anonymity.
 - **Debate mode** — after cross-review, every Contested or Disputed finding goes back to its raiser to **defend, amend, or withdraw**, and the disputing judges re-vote. Exactly one rebuttal round, then the final tally.
-- **Chair verdict scale** — the chair closes with 3–5 hard questions and one parseable line: `VERDICT: Ship it | Fix these first | Fundamental rethink`.
 - **Claude in the council** — Claude adds its own fresh review to the bundle so the bench ranks and adjudicates it. Claude is *judged* but never votes or chairs, so the verdict stays independent.
 
 The critic and lens methodologies are adapted from the `/critic` and `/debate` agents in [John Renaldi's product-kit](https://github.com/jrenaldi79/plugin-marketplace) (MIT); the briefing boilerplate lives in [`skills/second-opinion/SEAT-BRIEFS.md`](./skills/second-opinion/SEAT-BRIEFS.md).
@@ -192,7 +191,7 @@ Every path delivers the MCP server and both skills. They differ in what else you
 
 #### With npm — recommended
 
-The canonical path, and the one that gets you the full interactive experience (needs [Node.js](https://nodejs.org) ≥ 18):
+The canonical path, and the one that gets you the full interactive experience (needs [Node.js](https://nodejs.org) ≥ 22.12):
 
 ```bash
 npm install -g amicus
@@ -202,7 +201,7 @@ This is the path to pick unless you specifically want the plugin's slash command
 
 #### With the install script
 
-Same result as npm, one command — macOS, Linux, or Windows (needs [Node.js](https://nodejs.org) ≥ 18):
+Same result as npm, one command — macOS, Linux, or Windows (needs [Node.js](https://nodejs.org) ≥ 22.12):
 
 ```bash
 # macOS / Linux
@@ -333,7 +332,7 @@ Everything you need before your first run, and what's optional.
 
 **Runtime**
 
-- **Node.js ≥ 18** — `node --version` to check. This is the only hard runtime prerequisite.
+- **Node.js ≥ 22.12** — `node --version` to check. This is the only hard runtime prerequisite.
 - **An active Claude Code or Cowork session** — Amicus is orchestrated by Claude; it is not a standalone chatbot.
 
 **Install path & the git toolchain**
@@ -441,7 +440,7 @@ $ amicus status demo123 --json
   "taskId": "demo123",
   "status": "complete",
   "elapsed": "5m 0s",
-  "version": "4.6.2",
+  "version": "4.6.3",
   "model": "google/gemini-2.5-flash",
   "phase": "terminal"
 }

package/docs/ROADMAP.md

@@ -7,10 +7,10 @@ major jump, gated on funding/cofounder. The observability arc is split so the **
 first (v4.3)** and the **Electron "Council Workspace" (v4.4)** rides on top of it. `--dry-run` cost
 preview dropped to the backlog.
 
-Amicus is at **v4.6.1** (tagged 2026-08-03). Each 4.x rev below leads with the benefit, not the
+Amicus is at **v4.6.2** (tagged 2026-08-05). Each 4.x rev below leads with the benefit, not the
 plumbing.
 
-**Status:** v4.0 through **v4.6.1** have **shipped** — everything down to the v4.7 heading is a
+**Status:** v4.0 through **v4.6.2** have **shipped** — everything down to the v4.7 heading is a
 record of what landed, not a plan. **v4.7 (composition + tagging + GUI ergonomics) is the next
 planned rev** — that scope carried the number v4.6 here until the degrade-announcement-invariant
 milestone took the v4.6.0 release (2026-08-02); renumbered, not descoped. v5.0 remains
@@ -140,6 +140,35 @@ more chance": the once-only Stage-1 retry with `Recovered:` heals, ruled heal-fi
 the **MCP update notice**, the `rebuildElectron` hint deletion, and the **fold-back-corrected
 MODEL-NOTES seed** (PR #93). Ninth consecutive first-attempt publish.
 
+### v4.6.1 / v4.6.2 — the field-report five *(patch train)* — ✅ SHIPPED v4.6.2, 2026-08-05
+Field-report-driven hardening in five sequential PRs plus one ruling follow-up: the
+`ANTHROPIC_BASE_URL` diagnosis pair (doctor row + host-form normalization) and stored-alias
+drift warning (#95); doctor-suite hermeticity (#96); the no-output backstop —
+`AMICUS_NO_OUTPUT_BACKSTOP_MS`, legs that produce nothing fail fast with a real reason (#99);
+`models --check --live` — one quiet paid wave proves stored aliases actually SERVE (#100);
+Workspace dead-seat rows — an announced-dead seat renders on the seats panel, live mid-poll
+after the owner's ruling, blind-masked (#102, #103); chair-attempt records — the fallback walk
+is diagnosable from `run.json` (`chairAttempts[]`, #105). The v4.6 announcement invariant now
+reaches the surface users watch and the artifact they keep.
+
+### v4.6.3 — the post-train sweep *(patch)* — ✅ SHIPPED v4.6.3, 2026-08-05
+A four-PR correction patch: the models audit stops crying wolf, the seats panel stops being
+fooled, and a handful of proven small defects land with tests.
+- **Audit routing-choice + fable direct route** — `models --check` stops flagging deliberate
+  gateway-only routes like `gpt-pro` STALE (no more harmful downgrade suggestion); fable gains an
+  authored `anthropic/claude-fable-5` route and routes direct-first with an Anthropic key — #107
+- **Role-aware, old-run-resilient dead-seat rows + openRun guard** — a model dead as critic but
+  alive as chair now renders its dead row; pre-v4.6 runs render dead rows from
+  `verdict.degrades`/`seatLoss.deadBenchSeats`; the third F09-class stale-reply hole
+  (`openRun`'s `get-run` reply) closes — #108
+- **Valueless `-o` + save-shadow notice + metadata tmp sweep** — a trailing bare `-o`/`--out`
+  now errors instead of orphaning a tmp file; `council save` announces when it shadows a
+  built-in bench; `doctor --fix` sweeps orphaned `metadata.json` tmp files — #109
+- **Registry body assert + Node-floor truth sweep + `makeBaseDeps()`** — the MCP-Registry
+  skip-check verifies the version *and* status it trusts, fail-toward-publish on every other
+  path; the README/install scripts/doctor all agree on the real Node ≥22.12 floor; eleven
+  duplicated doctor test fixtures consolidate into one factory — #110
+
 ## v4.7 — "Compose your councils" *(renumbered from v4.6, 2026-08-02 — spec + fresh plan at kickoff per the anti-rot rule)*
 **Benefit:** councils chain — generate → critique → refine with no manual copy-paste — and history
 becomes navigable.
@@ -148,8 +177,8 @@ becomes navigable.
 - **Session/wave tagging + `--search` + grouped history** (F8) *(S–M)*
 - **GUI power ergonomics** (F10: focus-follows fold hotkey, distinguishable window titles, tiling
   presets) *(S each)*
-- Deferred-item candidates per the v4.5 design doc's §8: RN-2, TST-1/TST-2, REL-2, CA-4, LC-5,
-  remainder of TST-7
+- Deferred-item candidates per the v4.5 design doc's §8: RN-2, TST-1/TST-2, REL-2, CA-4
+  (repair-solos half — the failed-chair half shipped v4.6.2), remainder of TST-7
 - **README + docs update** *(S)*
 > The 2026-07-19 combined spec (`2026-07-19-v4.5-policy-packs-composition-design.md`) holds the
 > approved chaining/tagging/F10 design detail and is the primary input to the v4.7 brainstorm; it

package/docs/configuration.md

@@ -247,9 +247,11 @@ Everything lives under `~/.config/amicus/` (`getConfigDir()` in `src/utils/confi
 metadata) use an atomic write: a temp file named `.<target>.<pid>.<random>.tmp` is written
 alongside the target, then renamed into place. A process killed between the write and the rename
 leaves an orphaned `.tmp` file behind forever — harmless, but it accumulates. `amicus doctor --fix`
-sweeps orphaned `sessions-index.json.*.tmp` files (only ones older than 60 seconds, so a live
-writer's in-flight tmp file is never touched); `amicus doctor` (without `--fix`) just reports the
-count.
+sweeps orphaned `sessions-index.json.*.tmp` files and, per-session, orphaned `.metadata.json.*.tmp`
+files (the B09 class, ~30 write sites — both `<taskId>/` and `<taskId>/subagents/<id>/` levels
+under the current project's `.claude/amicus_sessions/`); both sweeps only remove files older than
+60 seconds, so a live writer's in-flight tmp file is never touched. `amicus doctor` (without
+`--fix`) just reports the counts.
 
 ### Session storage
 

package/docs/usage.md

@@ -381,6 +381,12 @@ amicus models --check --live  # + probe every stored alias with a real leg (spen
 
 `amicus models --check` exits with the **number of stale aliases** (capped at 100) and prints same-vendor replacement suggestions for each, so it drops cleanly into CI.
 
+**Gateway-only routes.** A curated alias whose direct form is *derived* from its OpenRouter route (rather than
+authored) is not reported STALE when that direct form is missing from the vendor's
+direct namespace while the OpenRouter route still serves — a gateway-only route with
+no direct sibling is a routing choice, not staleness. Deliberately gateway-only
+entries (e.g. `gpt-pro`) are annotated as such and are never offered a retarget.
+
 **Drifted aliases.** `--check` (and the `doctor` aliases row) also flags **`DRIFTED:`** stored aliases — a stored alias whose target is still catalog-listed but no longer matches any route its family currently resolves to (the v4.6.1 `gemini` release-gate class, where `doctor` stayed green while the model behind it had moved on). Each drift line prints the exact `amicus setup --add-alias <alias>=<current>` refresh command. Drift is informational only — unlike stale aliases, it never changes the exit code.
 
 **Live probe (`--check --live`).** Presence in the catalog is not proof of service — a stored alias can point at a model id the catalog still lists but the provider has quietly stopped serving (the v4.6.1 `gemini` incident). `--check` alone can't see that; `--live` can, by actually asking. Scope is **stored aliases only** (`amicus setup --add-alias`) — curated defaults follow the catalog by construction and have no "was it actually served" question for a live probe to answer. **This spends real money — one tiny leg per stored alias** — every probed alias gets one ordinary engine leg on a single quiet fan-out wave, with a real session dir and a real spend-ledger row, exactly as if you'd run it yourself.
@@ -462,7 +468,7 @@ $ amicus status demo123 --json
   "taskId": "demo123",
   "status": "complete",
   "elapsed": "5m 0s",
-  "version": "4.6.2",
+  "version": "4.6.3",
   "model": "google/gemini-2.5-flash",
   "phase": "terminal"
 }
@@ -514,7 +520,7 @@ Runs every check below, in order, and prints a ✓/⚠/✗ line for each plus a
 
 | Check | What it verifies | Can fail as |
 |---|---|---|
-| `node` | Node.js ≥ 18 | error |
+| `node` | Node.js ≥ 22.12 | error |
 | `config-dir` | The resolved config directory | *(always ok)* |
 | `keys` | At least one cloud-vendor key configured | error |
 | `default-model` | Your default model alias resolves | error |
@@ -528,13 +534,14 @@ Runs every check below, in order, and prints a ✓/⚠/✗ line for each plus a
 | `mcp` | Amicus is registered as an MCP server in Claude Code | warn |
 | `mcp-legacy` | No duplicate legacy `sidecar` MCP entry survives alongside `amicus` | warn |
 | `sessions-index-tmp` | No orphaned `sessions-index.json.*.tmp` files | warn |
+| `session-metadata-tmp` | No orphaned per-session `.metadata.json.*.tmp` files (the B09 class) | warn |
 | `openrouter-credit` | Remaining OpenRouter credit (skipped — reports `ok` — when no OpenRouter key is set) | warn |
 | `local-providers` **(v4.2)** | Every provider in `config.providers` is reachable | warn |
 | `project-root` | Your cwd looks like a real project, not an app/install dir | warn |
 
 **`local-providers`** probes every configured local provider (2s timeout each) the same way `amicus provider test` does, and reports per-id reachability in one line, e.g. `ollama: 3 models @ http://127.0.0.1:11434/v1; my-vllm: unreachable @ http://127.0.0.1:8000/v1`. No providers configured at all is a plain `ok` ("none configured") — this check can never fail your doctor run outright, only warn: a napping `ollama serve` isn't treated as broken setup.
 
-`--fix` self-heals four of the checks above in place: reprovisions Electron, copies the OpenCode engine into a broken npx-cache install, removes a duplicate legacy MCP entry, and sweeps orphaned session-index tmp files (only ones older than 60s). It does **not** start a local server for you — `local-providers` stays a warning until you start the server yourself.
+`--fix` self-heals five of the checks above in place: reprovisions Electron, copies the OpenCode engine into a broken npx-cache install, removes a duplicate legacy MCP entry, sweeps orphaned session-index tmp files, and sweeps orphaned per-session metadata tmp files (both tmp sweeps only ones older than 60s). It does **not** start a local server for you — `local-providers` stays a warning until you start the server yourself.
 
 Exit code is `1` if anything is `error`, else `0` (same rule drives `--json`'s `ok` field).
 

package/electron/workspace-ui/live-model.js

@@ -37,9 +37,17 @@
     return (state.visible && state.focused) ? 1500 : 5000;
   }
 
+  /** True when `status` is a terminal run status. The single consumption
+   * point for TERMINAL_STATUSES membership (v4.6.3 PR2 dedup) — the array
+   * itself stays exported and byte-identical to src/workspace/run-detail.js
+   * (drift-pinned). */
+  function isTerminal(status) {
+    return TERMINAL_STATUSES.indexOf(status) !== -1;
+  }
+
   /** Blind default per spec resolved Q2: ON live, OFF terminal. */
   function defaultBlind(status) {
-    return TERMINAL_STATUSES.indexOf(status) === -1;
+    return !isTerminal(status);
   }
 
   function dash(v) {
@@ -130,20 +138,56 @@
    * seatCells): an identity that is not carefully matched silently
    * duplicates or overwrites instead of failing loud.
    *
+   * Role-aware D6 (v4.6.3 PR2, spec D3): a bare model match used to suppress
+   * regardless of what the LIVE row's role was — so a model that died as
+   * critic but whose chair-fallback walk happened to land on that same alias
+   * (and succeeded, producing a live `role: 'chair'` cost row) silently
+   * erased the dead-critic row it was never a replacement for (spec §5, the
+   * PR 102 rider). Candidates now carry a `role` (`'critic'` via alias equality
+   * with `runMeta.critic` — mirroring `deriveSeatLoss`, verdict.js:72 — or
+   * `null`), and only REVIEWING-role live legs (`seat`/`critic`/`lens:*`)
+   * suppress at all; a `'critic'` candidate is cleared only by a live
+   * CRITIC-role leg for that alias, never by a chair/judge/rebuttal/revote
+   * row landing on the same model. Hidden dependency: the recovered-critic
+   * suppression below (`byRole[alias + '|critic']`) relies on `roleFor`'s
+   * critic branch (src/council/run-stages.js), which only fires when lenses
+   * are absent — safe today only because --critic and --lenses are mutually
+   * exclusive (cli-handlers-council-run.js:196); if that exclusion ever
+   * loosens, a healed critic on a lens run would carry role 'seat' and this
+   * suppression would render a ghost dead row for it.
+   *
+   * Old-run resilience (v4.6.3 PR2, spec D4): pre-`degrades[]` runs (v4.5.2)
+   * carry the BENCH half of a seat loss only in `seatLoss.deadBenchSeats`
+   * (string[] of aliases, verdict.js deriveSeatLoss) — `degrades[]` never
+   * existed on either doc for these runs. Consumed after the critic backstop,
+   * candidates get `role: null` (deadBenchSeats carries no critic/bench
+   * distinction beyond what `criticRequested` already covers above) and flow
+   * through the same `seen`-keyed dedup and role-aware suppression as every
+   * other candidate — deriveSeatLoss does not dedup its own array, so `seen`
+   * is what keeps a repeated alias (or one also named by a real degrade
+   * record) from rendering twice.
+   *
    * @param {Array<object>} degrades  run.json's `degrades[]` (may be absent)
-   * @param {?object} seatLoss  verdict.json's `seatLoss` (may be absent)
+   * @param {?object} seatLoss  verdict.json's `seatLoss` (may be absent) —
+   *   `criticRequested`/`criticSeated` back the critic candidate above,
+   *   `deadBenchSeats` (string[] of aliases) feeds the bench candidates below
    * @param {Array<{model: string}>} liveSeats  seatsFromRunStats(...)'s output
    *   (or any seat list keyed the same way — the live seat map)
-   * @returns {Array<{model: string, statusText: string}>}
+   * @param {?{critic: ?string}} runMeta  run.critic (alias, or null/absent
+   *   when no critic was requested) — degrade records carry no role field, so
+   *   this is the ONLY way a degrade-sourced candidate is identified as critic
+   * @returns {Array<{model: string, statusText: string, role: ?string}>}
    */
-  function deadSeats(degrades, seatLoss, liveSeats) {
+  function deadSeats(degrades, seatLoss, liveSeats, runMeta) {
+    var critic = runMeta && runMeta.critic ? runMeta.critic : null;
     var seen = {};
     var order = [];
-    function add(model, retried) {
+    function add(model, retried, role) {
       if (!model || seen[model]) { return; }
       seen[model] = true;
       order.push({
         model: model,
+        role: role || null,
         statusText: retried ? 'did not review — retried once' : 'did not review',
       });
     }
@@ -152,16 +196,36 @@
       if (d.channel !== 'dead-leg' && d.channel !== 'dead-wave') { return; }
       var data = d.data || {};
       var retried = !!(data.retryWaveId || data.firstFailure);
+      // Critic identification mirrors deriveSeatLoss (verdict.js): alias
+      // equality with run.critic — degrade records carry no role field.
       if (d.channel === 'dead-leg') {
-        add(data.seat, retried);
+        add(data.seat, retried, critic && data.seat === critic ? 'critic' : null);
       } else {
-        (data.models || []).forEach(function (m) { add(m, retried); });
+        (data.models || []).forEach(function (m) {
+          add(m, retried, critic && m === critic ? 'critic' : null);
+        });
       }
     });
     if (seatLoss && seatLoss.criticRequested && !seatLoss.criticSeated) {
-      add(seatLoss.criticRequested, false);
+      add(seatLoss.criticRequested, false, 'critic');
     }
-    var live = {};
+    if (seatLoss) {
+      // Pre-degrades[] era (v4.5.2): the bench half of a seat loss lives
+      // only here. Alias strings; deriveSeatLoss does not dedup — `seen`
+      // absorbs repeats and degrade-sourced duplicates.
+      (seatLoss.deadBenchSeats || []).forEach(function (m) { add(m, false, null); });
+    }
+    // Role-aware D6 (v4.6.3 PR2): only REVIEWING-role live legs suppress —
+    // a chair/judge/rebuttal/revote row must not hide a dead reviewer, and
+    // a dead-critic candidate is cleared only by a live CRITIC leg. A null
+    // role is NOT reviewing: counting it would suppress silently, the exact
+    // class the announcement invariant forbids.
+    function isReviewing(role) {
+      return role === 'seat' || role === 'critic' ||
+        (typeof role === 'string' && role.indexOf('lens:') === 0);
+    }
+    var reviewing = {};
+    var byRole = {};
     // ⚠️ Fable review (PR4b fix wave): same F34/F36 alias-selection seatCells already uses
     // (`seat.modelInput || seat.model`, above) — a LIVE payload seat's `model` is the RESOLVED
     // executable id, not the alias a degrade record names; `modelInput` carries the alias.
@@ -170,14 +234,23 @@
     // so D6 failed to suppress it — both rows rendered until the stage boundary dropped the
     // errored row. Terminal-path cost rows (seatsFromRunStats) carry no `modelInput` at all and
     // are already alias-only, so `|| s.model` leaves that path unchanged.
-    (liveSeats || []).forEach(function (s) { live[s.modelInput || s.model] = true; });
-    return order.filter(function (s) { return !live[s.model]; });
+    (liveSeats || []).forEach(function (s) {
+      if (!isReviewing(s.role)) { return; }
+      var alias = s.modelInput || s.model; // F36: alias space, never resolved ids
+      reviewing[alias] = true;
+      byRole[alias + '|' + s.role] = true;
+    });
+    return order.filter(function (s) {
+      if (s.role === 'critic') { return !byRole[s.model + '|critic']; }
+      return !reviewing[s.model];
+    });
   }
 
   // ⚠️ DE-ROT (F41): STAGE_LABELS is exported so applyLive() can label post-open stages.
   var api = { pollDelay: pollDelay, seatCells: seatCells, seatsFromRunStats: seatsFromRunStats,
     deadSeats: deadSeats,
-    defaultBlind: defaultBlind, dash: dash, TERMINAL_STATUSES: TERMINAL_STATUSES, STAGE_LABELS: STAGE_LABELS };
+    defaultBlind: defaultBlind, isTerminal: isTerminal, dash: dash,
+    TERMINAL_STATUSES: TERMINAL_STATUSES, STAGE_LABELS: STAGE_LABELS };
   if (typeof module !== 'undefined' && module.exports) { module.exports = api; }
   if (typeof window !== 'undefined') { window.AmicusLive = api; }
 })();

package/electron/workspace-ui/workspace-app.js

@@ -67,6 +67,11 @@
     state.runId = runId;
     state.debate = null;
     return invoke('workspace:get-run', runId).then(function (detail) {
+      // F09 guard (v4.6.3 PR2): a reply for a run the user has since
+      // navigated away from must never overwrite the run now open. Guard on
+      // runId movement ONLY — a same-run re-open (the live loop's terminal
+      // refresh) must still apply its fresher reply.
+      if (state.runId !== runId) { return; }
       state.detail = detail;
       // ⚠️ DE-ROT (F38): debate.json is the re-vote index the matrix drill-in needs — fetched
       // once per run-open (never per render), fire-and-forget. An aborted or cost-ceiling
@@ -146,7 +151,7 @@
     P.renderVerdictPanel();
     R.renderCost($('cost-body'), d.derived.cost, state.blind, labelOf);
     P.wireLazyPanels();
-    var isTerminal = window.AmicusLive.TERMINAL_STATUSES.indexOf(d.run.status) !== -1;
+    var isTerminal = window.AmicusLive.isTerminal(d.run.status);
     $('abort-btn').hidden = isTerminal;
     V.startLiveLoop();
   }

package/electron/workspace-ui/workspace-render.js

@@ -36,6 +36,12 @@
     return el('span', { className: 'chip ' + (kind || ''), title: text }, [text]);
   }
 
+  /** td className for seat-table cell index i — one source for the three
+   * call sites (renderSeats create + update, renderDeadSeatRows). */
+  function seatCellClass(i) {
+    return i >= 4 && i <= 6 ? 'num' : (i === 8 ? 'stalled-flag' : '');
+  }
+
   /** Dual-name display flip (blind mode). pair = {model, label}. */
   function display(pair, blind) {
     if (!pair) { return '—'; }
@@ -197,7 +203,7 @@
       var cells = window.AmicusLive.seatCells(view, blindOn, labelOf);
       if (!row) {
         row = el('tr', { dataset: { key: key } }, cells.map(function (c, i) {
-          return el('td', { className: i >= 4 && i <= 6 ? 'num' : (i === 8 ? 'stalled-flag' : '') }, [c]);
+          return el('td', { className: seatCellClass(i) }, [c]);
         }));
         tbody.appendChild(row);
         return;
@@ -205,7 +211,7 @@
       cells.forEach(function (c, i) {
         var td = row.children[i];
         if (td && td.textContent !== c) { td.textContent = c; }
-        if (td) { td.className = i >= 4 && i <= 6 ? 'num' : (i === 8 ? 'stalled-flag' : ''); }
+        if (td) { td.className = seatCellClass(i); }
       });
     });
     // RN-11 (v4.5): the keyed update added and removed rows but never MOVED
@@ -282,6 +288,6 @@
     el: el, chip: chip, display: display, relTime: relTime, renderRunList: renderRunList,
     renderHeaderChips: renderHeaderChips, renderGauge: renderGauge, renderStageRail: renderStageRail,
     renderSeats: renderSeats, renderBanner: renderBanner, renderCost: renderCost,
-    renderProseSections: renderProseSections,
+    renderProseSections: renderProseSections, seatCellClass: seatCellClass,
   };
 })();

package/electron/workspace-ui/workspace-seats.js

@@ -48,13 +48,20 @@
     var tbody = A.$('seats-body');
     window.AmicusRender.renderSeats(tbody, seats, A.state.blind, A.labelOf);
     var seatLoss = d.verdict && d.verdict.seatLoss;
-    var dead = window.AmicusLive.deadSeats(d.run.degrades, seatLoss, seats);
+    var runMeta = { critic: (d.run && d.run.critic) || null };
+    // Source-selection (v4.6.3 PR2, spec D4): run-degrade.js swallows checkpoint failures, so
+    // verdict.json can carry degrade records run.json's own checkpoint lost — fall back to it
+    // ONLY when run.degrades is empty/absent. A fallback, never a union: both docs can carry
+    // records for the SAME run, and the persisted run.json copy is authoritative when present.
+    var deg = (d.run && d.run.degrades && d.run.degrades.length) ? d.run.degrades
+      : ((d.verdict && d.verdict.degrades) || []);
+    var dead = window.AmicusLive.deadSeats(deg, seatLoss, seats, runMeta);
     renderDeadSeatRows(tbody, dead, A.state.blind, A.labelOf);
   }
 
   /**
    * Paints the dead-seat rows appended after live rows. Deliberately NOT
-   * folded into workspace-render.js's renderSeats (287/300 — must not grow)
+   * folded into workspace-render.js's renderSeats (293/300 — must not grow)
    * and NOT run through its keyed diff: dead rows carry no per-tick-changing
    * field, so a full rebuild every call is correct and cheap, and renderSeats
    * just above already self-cleans any PRIOR dead row as an unrecognized
@@ -74,7 +81,7 @@
   function renderDeadSeatRows(tbody, dead, blindOn, labelOf) {
     (dead || []).forEach(function (seat) {
       var cells = window.AmicusLive.seatCells(
-        { model: seat.model, status: seat.statusText, stalled: false }, blindOn, labelOf);
+        { model: seat.model, role: seat.role, status: seat.statusText, stalled: false }, blindOn, labelOf);
       // Fix wave 2 (smoke-caught, GUI smoke on real degraded run 12c96b6b): dead seats never
       // produce a review, so state.labelByModel (built from the run's names derivation — models
       // that DID review) never carries them; seatCells' own `blindOn && label ? label : alias`
@@ -88,7 +95,7 @@
         { className: 'seat-dead', dataset: { key: 'dead:' + seat.model } },
         cells.map(function (c, i) {
           return window.AmicusRender.el('td',
-            { className: i >= 4 && i <= 6 ? 'num' : (i === 8 ? 'stalled-flag' : '') }, [c]);
+            { className: window.AmicusRender.seatCellClass(i) }, [c]);
         }));
       tbody.appendChild(row);
     });
@@ -105,7 +112,15 @@
     var A = window.AmicusApp;
     var d = A.state.detail;
     var seatLoss = d && d.verdict ? d.verdict.seatLoss : null;
-    var dead = window.AmicusLive.deadSeats(live.degrades, seatLoss, live.seats || []);
+    var runMeta = { critic: (d && d.run && d.run.critic) || null };
+    // Source-selection (v4.6.3 PR2, spec D4), live-path twin of renderSeatsPanel's fallback
+    // above: the tick's own live.degrades wins when non-empty; state.detail.verdict.degrades is
+    // usually absent mid-run (verdict.json doesn't exist until the run finishes) — fine, this
+    // branch only matters for the rare same-run reopen where a prior terminal fetch already
+    // populated state.detail.verdict.
+    var deg = (live.degrades && live.degrades.length) ? live.degrades
+      : ((d && d.verdict && d.verdict.degrades) || []);
+    var dead = window.AmicusLive.deadSeats(deg, seatLoss, live.seats || [], runMeta);
     renderDeadSeatRows(A.$('seats-body'), dead, A.state.blind, A.labelOf);
   }
 

package/electron/workspace-ui/workspace-verbs.js

@@ -66,7 +66,7 @@
     var A = window.AmicusApp;
     stopLiveLoop();
     var d = A.state.detail;
-    if (!d || !d.run || window.AmicusLive.TERMINAL_STATUSES.indexOf(d.run.status) !== -1) { return; }
+    if (!d || !d.run || window.AmicusLive.isTerminal(d.run.status)) { return; }
     var epoch = A.state.liveEpoch; // F42: stopLiveLoop() above just bumped it; this chain owns it
     var tick = function () {
       // F42: pin the id PER TICK and SEND the pinned id — matching the reply against

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "amicus",
-  "version": "4.6.2",
+  "version": "4.6.3",
   "mcpName": "io.github.BourbonDog/amicus",
   "description": "Multi-model LLM Council + parallel AI window for Claude Code. Run structured council reviews across Gemini, GPT, DeepSeek and more — or fork a conversation to any model and fold the results back.",
   "keywords": [

package/src/cli-handlers-council.js

@@ -162,6 +162,15 @@ function runVerdict(args, useJson) {
         hint: 'pass a valid decisions.json array or omit --decisions' });
     }
   }
+  // R1 (v4.6.3): parseArgs records a valueless trailing -o/--out as boolean
+  // true (and --out= as ''). The boolean crashes writeVerdictAtomic mid-write
+  // (renameSync TypeError on a non-string path) leaving an orphaned
+  // true.tmp-<pid>; the empty string silently falls through to the default
+  // path. Name the flag and refuse both — the unknown-flag precedent.
+  if (args.out !== undefined && (typeof args.out !== 'string' || args.out === '')) {
+    return failJson(useJson, { code: ERROR_CODES.BAD_ARGS, message: '-o/--out requires a value',
+      hint: 'amicus council verdict <tally.json> [--decisions <decisions.json>] [-o|--out <out.json>]' });
+  }
   const outPath = args.out || './verdict.json';
   let verdict;
   try {

package/src/cli-handlers-doctor.js

@@ -80,10 +80,14 @@ function realDeps() {
     now: () => Date.now(),
     listSessionIndexTmpFiles: () => tmpSweep.listSessionIndexTmpFiles(), // B15
     unlinkSessionIndexTmp: (n) => tmpSweep.unlinkSessionIndexTmp(n),
+    listSessionMetadataTmpFiles: () => metaSweep.listSessionMetadataTmpFiles(), // D8
+    unlinkSessionMetadataTmp: (n) => metaSweep.unlinkSessionMetadataTmp(n),
   };
 }
 // B15: sweep logic in utils/session-index-tmp-sweep.js (mirrors mcp-legacy's split).
 const tmpSweep = require('./utils/session-index-tmp-sweep');
+// D8: per-session metadata.json sibling sweep — utils/session-metadata-tmp-sweep.js.
+const metaSweep = require('./utils/session-metadata-tmp-sweep');
 
 /** Run one guarded check; a thrown fn becomes an error line. */
 function guard(id, name, fn) {
@@ -108,10 +112,10 @@ async function runDoctorChecks(depsOverride = {}) {
   const checks = [];
 
   checks.push(guard('node', 'Node.js', () => {
-    const major = parseInt(String(d.nodeVersion).replace(/^v/, '').split('.')[0], 10);
-    return major >= 18
+    const [maj, min] = String(d.nodeVersion).replace(/^v/, '').split('.').map(n => parseInt(n, 10));
+    return (maj > 22 || (maj === 22 && min >= 12))
       ? { id: 'node', name: 'Node.js', status: 'ok', message: d.nodeVersion, hint: null }
-      : { id: 'node', name: 'Node.js', status: 'error', message: `${d.nodeVersion} (need >=18)`, hint: 'Install Node 18 or newer from https://nodejs.org' };
+      : { id: 'node', name: 'Node.js', status: 'error', message: `${d.nodeVersion} (need >=22.12)`, hint: 'Install Node 22.12 or newer from https://nodejs.org' };
   }));
 
   checks.push(guard('config-dir', 'Config directory', () => (
@@ -201,6 +205,8 @@ async function runDoctorChecks(depsOverride = {}) {
 
   checks.push(guard('sessions-index-tmp', 'Session index tmp files', () => tmpSweep.evaluateSessionIndexTmpSweep(d)));
 
+  checks.push(guard('session-metadata-tmp', 'Session metadata tmp files', () => metaSweep.evaluateSessionMetadataTmpSweep(d)));
+
   // #43: OpenRouter credit/free-tier — warns (never errors); skipped when no key.
   checks.push(await guardAsync('openrouter-credit', 'OpenRouter credit', async () => {
     const values = d.readApiKeyValues() || {};

package/src/council/presets-cli.js

@@ -45,18 +45,22 @@ function runSave(name, modelsArg, useJson) {
       hint: 'each member must be a known alias (see `amicus models`) or a provider/model id containing "/"' });
   }
   const overwritten = !!getCouncil(name);
+  const shadowsBuiltin = listBuiltinCouncilNames().includes(name);
   const cfg = loadConfig() || {};
   if (!cfg.councils) { cfg.councils = {}; }
   cfg.councils[name] = members;
   saveConfig(cfg);
-  const doc = { ok: true, name, models: members, overwritten };
+  const doc = { ok: true, name, models: members, overwritten, shadowsBuiltin };
   process.stdout.write(useJson ? JSON.stringify(doc, null, 2) + '\n' : renderSave(doc));
   return 0;
 }
 
 function renderSave(doc) {
   const notice = doc.overwritten ? ' (overwritten)' : '';
-  return `Saved council '${doc.name}'${notice}: ${doc.models.join(', ')}\n` +
+  const shadow = doc.shadowsBuiltin
+    ? `  note: '${doc.name}' now shadows the built-in bench of the same name — 'amicus council list' shows both\n`
+    : '';
+  return `Saved council '${doc.name}'${notice}: ${doc.models.join(', ')}\n` + shadow +
     "  for full run configuration — chair, options, templates — see 'amicus pack'\n";
 }
 

package/src/council/run-chair.js

@@ -143,12 +143,14 @@ async function runChair(ctx, { packet, degrade, statsFn, isSignalled }) {
     emitStageStarted(o.runDir, o.runId, 'chair', null, o.follow);
     // Fallback chain (spec §4): retry same chair once → promote best
     // non-bench model from the ledger → give up (no Claude fallback headless).
-    let attempt = await attemptChair(o.chair, `${o.runId}-ch1`);
-    recordAttempt(attempt, `${o.runId}-ch1`, o.chair);
+    const waveId1 = `${o.runId}-ch1`;
+    let attempt = await attemptChair(o.chair, waveId1);
+    recordAttempt(attempt, waveId1, o.chair);
     if (isAbortExit(attempt.exitCode) || isSignalled()) { return bail(attempt.exitCode || isSignalled()); }
     if (!attempt.leg && !overBudget()) {
-      attempt = await attemptChair(o.chair, `${o.runId}-ch2`);
-      recordAttempt(attempt, `${o.runId}-ch2`, o.chair);
+      const waveId2 = `${o.runId}-ch2`;
+      attempt = await attemptChair(o.chair, waveId2);
+      recordAttempt(attempt, waveId2, o.chair);
       if (isAbortExit(attempt.exitCode) || isSignalled()) { return bail(attempt.exitCode || isSignalled()); }
     }
     if (attempt.leg) { actualChair = o.chair; }
@@ -157,8 +159,9 @@ async function runChair(ctx, { packet, degrade, statsFn, isSignalled }) {
       try { statsRows = statsFn(); } catch { /* no ledger yet */ }
       const fallback = pickFallbackChair(statsRows, o.models, o.chair);
       if (fallback) {
-        attempt = await attemptChair(fallback, `${o.runId}-ch3`);
-        recordAttempt(attempt, `${o.runId}-ch3`, fallback);
+        const waveId3 = `${o.runId}-ch3`;
+        attempt = await attemptChair(fallback, waveId3);
+        recordAttempt(attempt, waveId3, fallback);
         if (isAbortExit(attempt.exitCode) || isSignalled()) { return bail(attempt.exitCode || isSignalled()); }
         if (attempt.leg) { actualChair = fallback; }
       }

package/src/session-manager.js

@@ -22,6 +22,9 @@ const SESSION_STATUS = {
 /** Canonical session dir name — new sessions are written here. */
 const SESSIONS_DIR = 'amicus_sessions';
 
+/** Subagent sessions nest one level under their parent taskId dir. */
+const SUBAGENTS_DIR = 'subagents';
+
 /**
  * Get the canonical session directory path for a task (used for WRITES).
  * Spec Reference: §8.1 Session directory structure
@@ -256,7 +259,7 @@ function saveSummary(projectDir, taskId, summary) {
  * // Returns: '/path/to/project/.claude/amicus_sessions/abc123/subagents/subagent-xyz'
  */
 function getSubagentDir(projectDir, parentTaskId, subagentId) {
-  return path.join(getSessionDir(projectDir, parentTaskId), 'subagents', subagentId);
+  return path.join(getSessionDir(projectDir, parentTaskId), SUBAGENTS_DIR, subagentId);
 }
 
 /**
@@ -355,7 +358,7 @@ function getSubagentSession(projectDir, parentTaskId, subagentId) {
  * @returns {object[]} Array of sub-agent metadata
  */
 function listSubagents(projectDir, parentTaskId, filter = {}) {
-  const subagentsDir = path.join(getSessionDir(projectDir, parentTaskId), 'subagents');
+  const subagentsDir = path.join(getSessionDir(projectDir, parentTaskId), SUBAGENTS_DIR);
 
   if (!fs.existsSync(subagentsDir)) {
     return [];
@@ -412,6 +415,7 @@ module.exports = {
   getSessionDir,
   resolveExistingSessionDir,
   SESSIONS_DIR,
+  SUBAGENTS_DIR,
   SESSION_STATUS,
   // Sub-agent functions
   getSubagentDir,

package/src/utils/alias-audit.js

@@ -62,6 +62,12 @@ function idsByProvider(catalog) {
  * at least one live resolution from any other source. This ensures the
  * suggested `--add-alias` fix actually clears the warning, and prevents
  * permanently unclearable noise when the default openrouter route is live.
+ *
+ * A stale 'defaults' entry is additionally suppressed when it is the alias's
+ * DERIVED direct-form pin (per curated-models.directFormProvenance()) and the
+ * alias is either covered live from another source or declares gatewayOnly
+ * (v4.6.3 PR1, spec D2) — an AUTHORED defaults pin and user-config rows are
+ * never suppressed this way.
  * @param {Array<{alias,model,source}>} sources
  * @param {Array<{id:string}>} catalog
  */
@@ -75,11 +81,24 @@ function findStaleAliases(sources, catalog) {
   const covered = new Set(
     sources.filter(({ model }) => isLive(model) === true).map(({ alias }) => alias)
   );
+  // v4.6.3 PR1 (spec D2). Lazy-required so suites that doMock curated-models
+  // for other cases keep working; a stub without the accessor simply gets no
+  // suppression (fail-open toward reporting).
+  const cm = require('./curated-models');
+  const provenance = typeof cm.directFormProvenance === 'function' ? (cm.directFormProvenance() || {}) : {};
   return sources.filter(({ alias, model, source }) => {
     const ids = byProvider.get(model.split('/')[0]);
     if (!ids) { return false; } // provider unverifiable
     if (ids.has(model)) { return false; } // live
     if (source.startsWith('curated-route') && covered.has(alias)) { return false; }
+    // A 'defaults' pin that is the alias's DERIVED direct form: its absence
+    // from the direct namespace is a routing fact, not staleness, while the
+    // alias has live coverage (or declares gatewayOnly). The fix: suggestion
+    // this row would otherwise print is a retarget nobody should run — the
+    // 2026-08-05 release-gate false positive (v4.6.3 spec §3).
+    const prov = provenance[alias];
+    if (source === 'defaults' && prov && prov.directForm === 'derived' &&
+        (prov.gatewayOnly || covered.has(alias))) { return false; }
     return true;
   });
 }

package/src/utils/curated-models.js

@@ -68,7 +68,9 @@ const FAMILIES = [
 ];
 
 /**
- * Alias-only entries (no wizard quick pick); openrouter route only.
+ * Alias-only entries (no wizard quick pick). Every entry authors an
+ * openrouter route; entries whose vendor's direct API genuinely serves the
+ * model also author a direct route (claude/sonnet/haiku/fable).
  * Refreshed against the live catalog 2026-08-04.
  */
 const CARDLESS = [
@@ -77,7 +79,12 @@ const CARDLESS = [
   // gpt-5.5-pro ($30/$180 — still served, but expected to sunset with the
   // 5.5 line). `gpt-pro` tracks SOL while the `gpt` family tracks terra —
   // see the tier-semantics comment on the `gpt` family above.
-  { alias: 'gpt-pro', routes: { openrouter: 'openrouter/openai/gpt-5.6-sol-pro' } },
+  // gatewayOnly (owner ruling 2026-08-05, recorded in the v4.6.3 spec): the
+  // openrouter-only route is a deliberate routing choice — OpenAI's direct
+  // namespace does not serve gpt-5.6-sol-pro, so the DERIVED direct form
+  // must never be audited as stale and no direct pairing may be suggested.
+  { alias: 'gpt-pro', gatewayOnly: true,
+    routes: { openrouter: 'openrouter/openai/gpt-5.6-sol-pro' } },
   // codex: newest codex-specific model on OpenRouter (verified 2026-06-09).
   { alias: 'codex', routes: { openrouter: 'openrouter/openai/gpt-5.3-codex' } },
   { alias: 'claude', routes: { openrouter: 'openrouter/anthropic/claude-sonnet-5',
@@ -86,7 +93,11 @@ const CARDLESS = [
                                 anthropic: 'anthropic/claude-sonnet-5' } },
   { alias: 'haiku', routes: { openrouter: 'openrouter/anthropic/claude-haiku-4.5',
                               anthropic: 'anthropic/claude-haiku-4-5-20251001' } },
-  { alias: 'fable', routes: { openrouter: 'openrouter/anthropic/claude-fable-5' } },
+  // fable: direct route authored 2026-08-05 (owner ruling R2, v4.6.3 spec §3).
+  // Anthropic's /v1/models lists claude-fable-5 AND the direct route serves
+  // (live smoke wave 47278069) — the entry was OpenRouter-only at authoring.
+  { alias: 'fable', routes: { openrouter: 'openrouter/anthropic/claude-fable-5',
+                              anthropic: 'anthropic/claude-fable-5' } },
   { alias: 'qwen', routes: { openrouter: 'openrouter/qwen/qwen3.7-max' } },
   { alias: 'qwen-coder', routes: { openrouter: 'openrouter/qwen/qwen3-coder-next' } },
   { alias: 'qwen-flash', routes: { openrouter: 'openrouter/qwen/qwen3.6-flash' } },
@@ -160,7 +171,8 @@ function listCuratedRoutes() {
  * Vendors whose direct-API ids differ from OpenRouter's (dot vs. dash
  * versioning, distinct model names, etc.). NEVER derive a direct form for
  * these — derivation would emit the wrong (dot) id, or invent a direct id
- * for a model that is OpenRouter-only today (e.g. fable).
+ * for a model the direct API does not serve (fable was that case until its
+ * direct route was verified and authored, 2026-08-05).
  * Frozen so consumers can only read it (`.has()`) — a frozen Set still
  * supports lookups, it just can't be `.add()`/`.delete()`/`.clear()`-ed.
  */
@@ -200,6 +212,28 @@ function gatewayRoutesFor(vendorPath, obj) {
   return routes;
 }
 
+/**
+ * Per-alias provenance of the `direct` form in toGatewayRoutes(), for the
+ * auditors (alias-audit.js / gateway-route-audit.js): an AUTHORED direct
+ * form absent from its namespace is stale; a DERIVED one is a computed
+ * convenience whose absence is a routing fact, not staleness, while the
+ * authoring openrouter route is live. `gatewayOnly` mirrors an entry's
+ * explicit routing-choice annotation (owner-ruled): suppress derived-form
+ * findings unconditionally and never suggest a direct pairing.
+ * @returns {Object<string, {directForm: 'authored'|'derived'|'none', gatewayOnly: boolean}>}
+ */
+function directFormProvenance() {
+  const out = {};
+  const entryProv = (vendorPath, obj, gatewayOnly) => {
+    const direct = directFormFor(vendorPath, obj);
+    const directForm = !direct ? 'none' : (obj[vendorPath] ? 'authored' : 'derived');
+    return { directForm, gatewayOnly: gatewayOnly === true };
+  };
+  for (const f of FAMILIES) { out[f.alias] = entryProv(f.vendorPath, f.fallback, f.gatewayOnly); }
+  for (const e of CARDLESS) { out[e.alias] = entryProv(vendorOf(e.routes.openrouter), e.routes, e.gatewayOnly); }
+  return out;
+}
+
 /**
  * @returns {Object<string,{direct?: string, openrouter: string}>} alias →
  * per-gateway executable ids. Unlike `toDefaultAliases` (a single pinned
@@ -223,8 +257,9 @@ function toGatewayRoutes() {
  * disagree. It previously string-stripped `openrouter/` itself, which emitted
  * OpenRouter's dot ids for divergent vendors (`anthropic/claude-opus-4.8` —
  * the direct API only serves the dash form) and invented a bare direct id for
- * OpenRouter-only models (`fable`). Both made `amicus doctor` and `amicus
- * models --check` warn about the product's own shipped defaults.
+ * then-OpenRouter-only models (`fable`, direct-authored 2026-08-05). Both made
+ * `amicus doctor` and `amicus models --check` warn about the product's own
+ * shipped defaults.
  */
 function toDefaultAliases() {
   const out = {};
@@ -235,5 +270,6 @@ function toDefaultAliases() {
 }
 
 module.exports = {
-  getFamilies, toDefaultAliases, toCanonicalDefault, listCuratedRoutes, toGatewayRoutes, DIVERGENT_VENDORS
+  getFamilies, toDefaultAliases, toCanonicalDefault, listCuratedRoutes, toGatewayRoutes,
+  directFormProvenance, DIVERGENT_VENDORS
 };

package/src/utils/gateway-route-audit.js

@@ -30,7 +30,7 @@
 
 'use strict';
 
-const { toGatewayRoutes } = require('./curated-models');
+const { toGatewayRoutes, directFormProvenance } = require('./curated-models');
 const { classifyModel } = require('./model-classification');
 const { pairAcrossGateways } = require('./gateway-route-catalog');
 const { isDirectProvider } = require('./provider-registry');
@@ -70,19 +70,32 @@ function isAuthoritative(catalogInfo, id) {
  */
 function auditGatewayRoutes(catalogInfo) {
   const routes = toGatewayRoutes();
+  const provenance = directFormProvenance();
   const findings = [];
 
   for (const [alias, forms] of Object.entries(routes)) {
+    const prov = provenance[alias] || { directForm: 'none', gatewayOnly: false };
     for (const gateway of ['direct', 'openrouter']) {
       const id = forms[gateway];
       if (!id) { continue; }
-      if (classifyModel(id, gateway, catalogInfo) === 'invalid') {
-        findings.push({ alias, gateway, kind: 'stale', model: id });
+      if (classifyModel(id, gateway, catalogInfo) !== 'invalid') { continue; }
+      // v4.6.3 PR1 (spec D2): a DERIVED direct form is a computed convenience,
+      // not an authored claim. Its absence from the direct namespace is a
+      // routing fact — not staleness — while the authoring openrouter route
+      // is live, or when the entry declares gatewayOnly (an owner-ruled
+      // routing choice). An AUTHORED direct form absent from its namespace
+      // reports exactly as before.
+      if (gateway === 'direct' && prov.directForm === 'derived' &&
+          (prov.gatewayOnly ||
+           classifyModel(forms.openrouter, 'openrouter', catalogInfo) === 'valid')) {
+        continue;
       }
+      findings.push({ alias, gateway, kind: 'stale', model: id });
     }
 
     const vendor = vendorOf(forms.openrouter);
     if (!vendor || !isDirectProvider(vendor)) { continue; } // gateway-only vendor: no direct route ever possible
+    if (prov.gatewayOnly) { continue; } // declared routing choice: never suggest a direct pairing
     const token = bareSegment(forms.openrouter, vendor);
     if (!token) { continue; }
     const paired = pairAcrossGateways(vendor, token, catalogInfo); // Task-5 contract: bare segment only

package/src/utils/model-fetcher.js

@@ -8,17 +8,19 @@
 const https = require('https');
 
 /**
- * Hardcoded Anthropic models (no public listing endpoint). This is the
- * DIRECT-API floor only — Fable is OpenRouter-only (see curated-models.js
- * DIVERGENT_VENDORS / CARDLESS 'fable' entry, which has no `anthropic` route)
- * and must never appear here: classifyModel() returns 'valid' on a floor HIT
- * before it ever checks `authoritative`, so listing an OR-only model here
- * would mislabel a direct-API request for it as valid.
+ * Hardcoded Anthropic floor: the anthropic/ rows a KEYLESS user (or a
+ * failed live fetch) gets. Every id here must be one the direct API
+ * GENUINELY serves — classifyModel() returns 'valid' on a floor HIT before
+ * it ever checks `authoritative`, so a speculative row would mislabel a
+ * dead direct-API request as valid. (fable joined 2026-08-05 after live
+ * verification — /v1/models lists claude-fable-5 and a direct smoke leg
+ * served; v4.6.3 spec §3.)
  */
 const ANTHROPIC_MODELS = [
   { id: 'anthropic/claude-opus-5', name: 'Claude Opus 5', contextLength: null, pricing: null },
   { id: 'anthropic/claude-opus-4-8', name: 'Claude Opus 4.8', contextLength: null, pricing: null },
   { id: 'anthropic/claude-sonnet-5', name: 'Claude Sonnet 5', contextLength: null, pricing: null },
+  { id: 'anthropic/claude-fable-5', name: 'Claude Fable 5', contextLength: null, pricing: null },
   { id: 'anthropic/claude-haiku-4-5', name: 'Claude Haiku 4.5', contextLength: null, pricing: null },
   // Dated snapshot: the id Anthropic's /v1/models actually lists, and the
   // `haiku` direct route curated-models.js authors. Without it the floor

package/src/utils/remediation-hints.js

@@ -87,6 +87,20 @@ const REMEDIATION_HINTS = Object.freeze({
    */
   sweepSessionIndexTmp:
     'amicus doctor --fix  (sweeps orphaned .sessions-index.json.*.tmp files left by an interrupted write)',
+
+  /**
+   * Orphaned per-session metadata.json.*.tmp files (v4.6.3 PR3 Task 3 / D8):
+   * same producer shape as sweepSessionIndexTmp above, one level down — a
+   * kill between an atomic write's tmp-write and rename leaves a stray temp
+   * file in a session directory forever. `doctor --fix` sweeps files older
+   * than 60s (never a live writer's ms-lived tmp).
+   *
+   * Voice ruling (Christian, 2026-08-03, sweepSessionIndexTmp above): applies
+   * verbatim here — the cause is definitional, not a guess, so this hint also
+   * keeps its confident voice rather than the unverified-cause voice.
+   */
+  sweepSessionMetadataTmp:
+    'amicus doctor --fix  (sweeps orphaned .metadata.json.*.tmp files left by an interrupted write)',
 });
 
 module.exports = REMEDIATION_HINTS;

package/src/utils/session-metadata-tmp-sweep.js

@@ -0,0 +1,136 @@
+// src/utils/session-metadata-tmp-sweep.js
+'use strict';
+
+/**
+ * v4.6.3 PR3 Task 3 (D8): orphaned per-session metadata.json.*.tmp sweep for
+ * `amicus doctor --fix`.
+ *
+ * A kill between writeFileAtomic's tmp-write and rename (any of the ~30
+ * metadata.json write sites — session-manager.js, sidecar/session-finalize.js,
+ * etc.) leaves a stray `.metadata.json.<pid>.<hex>.tmp` file in a session
+ * directory forever (the B09 orphan class). This module lists and removes
+ * them; src/cli-handlers-doctor.js composes the result into a check line.
+ * Structurally mirrors utils/session-index-tmp-sweep.js (B15's sibling).
+ *
+ * Enumeration decision (cwd-scoped, index rejected — recorded at plan time,
+ * docs/superpowers/plans/2026-08-05-v463-pr3-cli-doctor-odds.md): this walks
+ * `<process.cwd()>/.claude/amicus_sessions/` — each taskId dir plus its
+ * `subagents/<id>/` children — rather than consulting sessions-index.json.
+ * `amicus doctor` is a per-project surface; the index is best-effort and can
+ * point at OTHER projects (issue #40's cross-project fallback exists exactly
+ * because the index can be stale). Trusting it here to decide which
+ * directories a --fix sweep may unlink from risks touching an unrelated
+ * project on stale data — the wrong failure direction for a destructive
+ * operation. Walking the cwd-scoped tree directly can only ever find/remove
+ * files under the project doctor is already running against.
+ *
+ * Symlink safety (Task 3 review carry, v4.6.3 PR3): every stat in this walk
+ * is lstatSync, never statSync — this module never follows symlinks. A
+ * symlinked taskId or subagents directory could otherwise be traversed and
+ * have files unlinked through the link, effectively outside the sessions
+ * root; lstat closes that off at zero cost.
+ */
+
+const fs = require('fs');
+const path = require('path');
+const HINTS = require('./remediation-hints');
+
+/** Files older than this survive to the next --fix, never a live writer's ms-lived tmp. */
+const AGE_THRESHOLD_MS = 60 * 1000;
+
+/** The cwd-scoped sessions root: <cwd>/.claude/amicus_sessions. */
+function sessionsRoot() {
+  const { SESSIONS_DIR } = require('../session-manager');
+  return path.join(process.cwd(), '.claude', SESSIONS_DIR);
+}
+
+/** True when `basename` is an orphaned metadata tmp file (not e.g. progress.json.*.tmp). */
+function isMetadataTmp(basename) {
+  return basename.startsWith('.metadata.json.') && basename.endsWith('.tmp');
+}
+
+/** List metadata tmp files directly inside `dir`, named relative to `root`. */
+function listTmpIn(dir, root) {
+  let entries;
+  try { entries = fs.readdirSync(dir); } catch { return []; }
+  return entries
+    .filter(isMetadataTmp)
+    .map((basename) => {
+      let mtimeMs = null;
+      try { mtimeMs = fs.lstatSync(path.join(dir, basename)).mtimeMs; } catch { /* raced away — skip below */ }
+      return { name: path.relative(root, path.join(dir, basename)), mtimeMs };
+    })
+    .filter((f) => f.mtimeMs !== null);
+}
+
+/**
+ * List orphaned per-session metadata.json.*.tmp files under the cwd-scoped
+ * sessions root, covering both `<taskId>/` and `<taskId>/subagents/<id>/`.
+ * @returns {Array<{name: string, mtimeMs: number}>}
+ */
+function listSessionMetadataTmpFiles() {
+  const { SUBAGENTS_DIR } = require('../session-manager');
+  const root = sessionsRoot();
+  let taskIds;
+  try { taskIds = fs.readdirSync(root); } catch { return []; }
+  const found = [];
+  for (const taskId of taskIds) {
+    const taskDir = path.join(root, taskId);
+    let stat;
+    try { stat = fs.lstatSync(taskDir); } catch { continue; }
+    if (!stat.isDirectory()) { continue; }
+    found.push(...listTmpIn(taskDir, root));
+
+    const subagentsDir = path.join(taskDir, SUBAGENTS_DIR);
+    let subIds;
+    try { subIds = fs.readdirSync(subagentsDir); } catch { continue; }
+    for (const subId of subIds) {
+      const subDir = path.join(subagentsDir, subId);
+      let subStat;
+      try { subStat = fs.lstatSync(subDir); } catch { continue; }
+      if (!subStat.isDirectory()) { continue; }
+      found.push(...listTmpIn(subDir, root));
+    }
+  }
+  return found;
+}
+
+/** Delete one orphaned tmp file by name (relative to the sessions root; never an absolute/caller path). */
+function unlinkSessionMetadataTmp(name) {
+  const root = sessionsRoot();
+  fs.unlinkSync(path.join(root, name));
+}
+
+/**
+ * Compose the doctor check line for the metadata tmp-orphan sweep. Pure
+ * decision logic (list/sweep side effects come in via `d`); src/cli-handlers-doctor.js
+ * wraps this in guard() the same way it wires the sibling sessions-index-tmp check.
+ * @param {{listSessionMetadataTmpFiles: () => Array<{name:string, mtimeMs:number}>,
+ *   fix?: boolean, now: () => number, unlinkSessionMetadataTmp: (name: string) => void}} d
+ */
+function evaluateSessionMetadataTmpSweep(d) {
+  const id = 'session-metadata-tmp'; const name = 'Session metadata tmp files';
+  const files = d.listSessionMetadataTmpFiles() || [];
+  if (files.length === 0) {
+    return { id, name, status: 'ok', message: '0 orphaned tmp files', hint: null };
+  }
+  if (!d.fix) {
+    return { id, name, status: 'warn', message: `${files.length} orphaned tmp file(s) — run with --fix`, hint: HINTS.sweepSessionMetadataTmp };
+  }
+  const nowMs = d.now();
+  const sweepable = files.filter((f) => (nowMs - f.mtimeMs) > AGE_THRESHOLD_MS);
+  let swept = 0;
+  for (const f of sweepable) {
+    try { d.unlinkSessionMetadataTmp(f.name); swept += 1; } catch { /* best-effort — report what we got */ }
+  }
+  const remaining = files.length - swept;
+  if (remaining === 0) {
+    const fixFields = swept > 0 ? { fixed: true, fixDetail: `swept ${swept} orphaned session-metadata tmp file(s)` } : {};
+    return { id, name, status: 'ok', message: `swept ${swept} orphaned tmp file(s)`, hint: null, ...fixFields };
+  }
+  return { id, name, status: 'warn', message: `swept ${swept}, ${remaining} remaining (too fresh or unremovable)`, hint: HINTS.sweepSessionMetadataTmp };
+}
+
+module.exports = {
+  AGE_THRESHOLD_MS, listSessionMetadataTmpFiles, unlinkSessionMetadataTmp, evaluateSessionMetadataTmpSweep,
+};