@lcv-ideas-software/cross-review 4.2.1 → 4.2.3

package/docs/model-selection.md

@@ -36,28 +36,29 @@ unavailable, the round retries on the same model or skips that peer
 hatch is an explicit operator override via `CROSS_REVIEW_<PROVIDER>_MODEL`
 env-var per host — a deliberate decision, never a silent downgrade.
 
-| Peer             | Pin                   | Override env-var                |
-| ---------------- | --------------------- | ------------------------------- |
-| OpenAI/Codex     | `gpt-5.5`             | `CROSS_REVIEW_OPENAI_MODEL`     |
-| Anthropic/Claude | `claude-opus-4-7`     | `CROSS_REVIEW_ANTHROPIC_MODEL`  |
-| Google/Gemini    | `gemini-2.5-pro`      | `CROSS_REVIEW_GEMINI_MODEL`     |
-| DeepSeek         | `deepseek-v4-pro`     | `CROSS_REVIEW_DEEPSEEK_MODEL`   |
-| xAI/Grok         | `grok-4-latest`       | `CROSS_REVIEW_GROK_MODEL`       |
-| Perplexity       | `sonar-reasoning-pro` | `CROSS_REVIEW_PERPLEXITY_MODEL` |
+| Peer             | Pin                      | Override env-var                |
+| ---------------- | ------------------------ | ------------------------------- |
+| OpenAI/Codex     | `gpt-5.5`                | `CROSS_REVIEW_OPENAI_MODEL`     |
+| Anthropic/Claude | `claude-opus-4-8`        | `CROSS_REVIEW_ANTHROPIC_MODEL`  |
+| Google/Gemini    | `gemini-3.1-pro-preview` | `CROSS_REVIEW_GEMINI_MODEL`     |
+| DeepSeek         | `deepseek-v4-pro`        | `CROSS_REVIEW_DEEPSEEK_MODEL`   |
+| xAI/Grok         | `grok-4.3`               | `CROSS_REVIEW_GROK_MODEL`       |
+| Perplexity       | `sonar-reasoning-pro`    | `CROSS_REVIEW_PERPLEXITY_MODEL` |
 
 Haiku and other low-capacity Anthropic models are intentionally excluded —
 the cross-review role requires advanced reasoning depth.
 
-Operator preference 2026-05-07: `gemini-2.5-pro` is the runtime default
-because under Google One AI Ultra subscription it carries 1k requests/day vs
-`gemini-3.1-pro-preview`'s 250 requests/day. Workspace policy: only
-`gemini-*-pro` variants ≥ 2.5 are permitted — no `*-flash` variants and no
-models below 2.5.
+Google's deprecation schedule lists `gemini-2.5-pro` for shutdown on
+2026-10-16 and recommends `gemini-3.1-pro-preview` as the replacement.
+Workspace policy remains: only `gemini-*-pro` variants >= 2.5 are permitted
+for this peer; no `*-flash` variants and no models below 2.5. Operators can
+still override the pin explicitly, but the default/canonical path follows the
+documented replacement.
 
-`GROK_API_KEY` is the canonical auth variable for xAI. The pinned
-`grok-4-latest` model uses automatic reasoning in this runtime, so the
-adapter omits the explicit `reasoning.effort` field automatically unless a
-future operator-approved pin explicitly supports that parameter.
+`GROK_API_KEY` is the canonical auth variable for xAI. The pinned `grok-4.3`
+model accepts explicit `reasoning.effort` values through `high`; the adapter
+clamps the shared effort scale so unsupported `xhigh`/`max` requests do not
+reach the wire.
 
 `PERPLEXITY_API_KEY` is the canonical auth variable for Perplexity Sonar.
 Sonar billing has a 3rd dimension: per-1000-requests fee that scales with
@@ -70,11 +71,17 @@ to skip search for the synthesis step.
 Cross-review is optimized for correctness over latency and cost. Provider adapters explicitly request thinking/reasoning where the official APIs support it:
 
 - OpenAI/Codex: Responses API with reasoning effort `xhigh` by default.
-- Anthropic/Claude: adaptive thinking with omitted thinking display plus `output_config.effort=xhigh` by default on Opus 4.7.
-- Google/Gemini: automatic thinking budget for the pinned Gemini 2.5 Pro model; Gemini 3.x thinking-level notes are historical and apply only to an explicit operator-approved override.
+- Anthropic/Claude: adaptive thinking with omitted thinking display plus `output_config.effort=xhigh` by default on Opus 4.8.
+- Google/Gemini: high thinking level for the pinned Gemini 3.1 Pro Preview
+  model; the adapter keeps the Gemini 3 thinking path explicit because this
+  peer is used for complex reasoning and coding review.
 - DeepSeek: `thinking.type=enabled` with `reasoning_effort=max` by default.
-- Grok: the pinned `grok-4-latest` model uses xAI automatic reasoning without
-  the explicit `reasoning.effort` field.
+- Grok: the pinned `grok-4.3` model accepts explicit `reasoning.effort`;
+  unsupported shared-scale values are clamped to the nearest supported value.
+- Perplexity: the pinned `sonar-reasoning-pro` model accepts an explicit
+  `reasoning_effort` enum (`minimal`/`low`/`medium`/`high`, `high` by default);
+  `clampEffortForPerplexity` narrows the shared effort scale into that range
+  (`none`/`minimal` → `minimal`; `xhigh`/`max` → `high`).
 
 ## Historical Documentation Refresh — 2026-05-05
 
@@ -85,18 +92,19 @@ above and enforced by `src/peers/model-selection.ts`.
 - OpenAI: GPT-5.5 is the current recommended frontier model for complex
   reasoning/coding, with Responses API reasoning effort values through `xhigh`
   and 1M context / 128K output.
-- Anthropic: Claude Opus 4.7 is the generally available complex-reasoning and
-  agentic-coding default; current docs expose 1M context and adaptive thinking.
-- Google Gemini: Gemini 3.1 Pro Preview was the advanced Gemini 3.1 option at
-  the time; Gemini 3 Pro Preview was deprecated/shut down and must stay out of
-  current pins and downgrade chains.
+- Anthropic: Claude Opus 4.8 supersedes Opus 4.7 as the current
+  complex-reasoning and agentic-coding default; current docs retain the same
+  regular price tier as 4.7.
+- Google Gemini: Gemini 3.1 Pro Preview is the documented replacement for
+  Gemini 2.5 Pro. Gemini 3 Pro Preview was deprecated/shut down and must stay
+  out of current pins and downgrade chains.
 - DeepSeek: DeepSeek-V4 exposes `deepseek-v4-pro` and `deepseek-v4-flash`;
   legacy `deepseek-chat` and `deepseek-reasoner` were scheduled for
   discontinuation on 2026-07-24 and must stay out of current pins and
   downgrade chains.
-- xAI Grok: historical Grok notes covered explicit-effort models that predate
-  the current `grok-4-latest` pin. Current runtime behavior is defined above:
-  omit explicit `reasoning.effort` for `grok-4-latest`.
+- xAI Grok: historical Grok notes covered aliases and explicit-effort models
+  that predate the current concrete `grok-4.3` pin. Current runtime behavior
+  is defined above: send clamped explicit `reasoning.effort` for `grok-4.3`.
 
 ## Important
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lcv-ideas-software/cross-review",
-  "version": "4.2.1",
+  "version": "4.2.3",
   "description": "API-first MCP server for multi-model cross-review with unanimous convergence gates.",
   "type": "module",
   "main": "dist/src/mcp/server.js",
@@ -24,11 +24,12 @@
     "dev": "tsx src/mcp/server.ts",
     "dashboard": "tsx src/dashboard/server.ts",
     "smoke": "tsx scripts/smoke.ts",
+    "provider-refresh-smoke": "tsx scripts/provider-refresh-smoke.ts",
     "runtime-smoke": "npm --registry=https://registry.npmjs.org run build && tsx scripts/runtime-smoke.ts",
     "runtime-default-smoke": "tsx scripts/runtime-default-smoke.ts",
     "api-streaming-smoke": "tsx scripts/api-streaming-smoke.ts",
     "release:verify-registry": "node scripts/verify-registry-dist.mjs",
-    "test": "npm --registry=https://registry.npmjs.org run build && npm --registry=https://registry.npmjs.org run smoke && npm --registry=https://registry.npmjs.org run runtime-smoke",
+    "test": "npm --registry=https://registry.npmjs.org run build && npm --registry=https://registry.npmjs.org run provider-refresh-smoke && npm --registry=https://registry.npmjs.org run smoke && npm --registry=https://registry.npmjs.org run runtime-smoke",
     "lint": "eslint . --max-warnings=0",
     "format": "prettier --write .",
     "format:check": "prettier --check .",
@@ -40,7 +41,7 @@
     "check": "npm --registry=https://registry.npmjs.org run format:check && npm --registry=https://registry.npmjs.org run lint && npm --registry=https://registry.npmjs.org run biome && npm --registry=https://registry.npmjs.org run typecheck"
   },
   "dependencies": {
-    "@anthropic-ai/sdk": "^0.97.1",
+    "@anthropic-ai/sdk": "^0.100.1",
     "@google/genai": "^2.5.0",
     "@modelcontextprotocol/sdk": "^1.29.0",
     "openai": "^6.38.0",
@@ -58,6 +59,7 @@
     "@types/node": "^25.9.1",
     "@types/proper-lockfile": "^4.1.4",
     "eslint": "^10.4.0",
+    "eslint-config-prettier": "^10.1.8",
     "prettier": "^3.8.3",
     "tsx": "^4.22.3",
     "typescript": "^6.0.3",
@@ -90,6 +92,7 @@
     "deepseek",
     "grok",
     "xai",
+    "perplexity",
     "api"
   ]
 }