github-router 0.3.175 → 0.3.176

package/README.md

@@ -83,7 +83,7 @@ Each persona is exposed both as a Claude Code subagent (callable via the `Task`
 
 | Subagent | Model | Endpoint | Effort tiers (default) |
 |---|---|---|---|
-| `codex-critic` | gpt-5.5 | `/v1/responses` | low \| medium \| high \| xhigh (xhigh) |
+| `codex-critic` | gpt-5.6-sol | `/v1/responses` | low \| medium \| high \| xhigh (xhigh) |
 | `codex-reviewer` | gpt-5.3-codex | `/v1/responses` | low \| medium \| high \| xhigh (xhigh) |
 | `gemini-reviewer` | gemini-3.1-pro-preview | `/v1/chat/completions` | low \| medium \| high (high) |
 | `opus-critic` | claude-opus-4-6 | `/v1/messages` | low \| medium \| high (high) |
@@ -116,7 +116,7 @@ The fastest path is the `codex` subcommand — it boots the proxy on a random po
 npx github-router@latest codex
 ```
 
-Defaults to `gpt-5.5`; falls back to `gpt-5.4` → `gpt-5.3-codex` → `gpt-5.2-codex` if your Copilot tier doesn't expose 5.5 yet. Override with `-m`:
+Defaults to `gpt-5.6-sol`; falls back to `gpt-5.5` → `gpt-5.4` → `gpt-5.3-codex` → `gpt-5.2-codex` if your Copilot tier doesn't expose 5.6-sol yet. Override with `-m`:
 
 ```sh
 npx github-router@latest codex -m gpt-5.3-codex
@@ -129,7 +129,7 @@ npx github-router@latest start --codex   # interactive launch-command generator
 # — or set env vars yourself —
 export OPENAI_BASE_URL="http://localhost:8787/v1"
 export OPENAI_API_KEY="dummy"
-codex --full-auto -m gpt-5.5
+codex --full-auto -m gpt-5.6-sol
 ```
 
 ---
@@ -168,7 +168,7 @@ Anthropic endpoints are only available under `/v1/messages`.
 | Model | /chat/completions | /responses | /v1/messages |
 |---|---|---|---|
 | gpt-4.1, gpt-4o | Yes | Yes | No |
-| gpt-5.5, gpt-5.4 | No | Yes | No |
+| gpt-5.6-sol, gpt-5.5, gpt-5.4 | No | Yes | No |
 | gpt-5.3-codex, gpt-5.2-codex | No | Yes | No |
 | claude-opus-4.8 (1M context as the single base slug) | Yes | No | Yes |
 | claude-opus-4.7-1m-internal (enterprise), claude-opus-4.7 | Yes | No | Yes |
@@ -237,7 +237,7 @@ github-router debug              Print diagnostic info
 The `claude` and `codex` subcommands accept all the shared flags below plus `-m`/`--model` to override the default model. Default models live in `src/lib/port.ts`:
 
 - `claude` → `claude-opus-4-8` (Anthropic dashed slug for UI compatibility; the proxy translates to Copilot's `claude-opus-4.8` — the single base slug already advertises 1M context via `max_context_window_tokens`, so no `-1m` sibling exists). Major.minor fallback chain: `claude-opus-4-7` → `claude-opus-4-6` → `claude-opus-4-5`.
-- `codex` → `gpt-5.5` → `gpt-5.4` → `gpt-5.3-codex` → `gpt-5.2-codex`
+- `codex` → `gpt-5.6-sol` → `gpt-5.5` → `gpt-5.4` → `gpt-5.3-codex` → `gpt-5.2-codex`
 
 Fallback chains fire only on the implicit-default path; explicit `-m`/`--model` is always respected as-is.
 

package/dist/browser-ext/manifest.json

@@ -2,7 +2,7 @@
   "manifest_version": 3,
   "name": "github-router browser bridge",
   "short_name": "gh-router-browser",
-  "version": "0.3.175",
+  "version": "0.3.176",
   "description": "Bridge between Claude (via github-router /mcp) and the browser. Implements tab control, navigation, clicks, form fill, downloads, screenshots, devtools eval. Blocks navigation to chrome://settings.",
   "key": "MIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEAqJElxuBlonBS3TVW9FJN0mGTtShB3L1hoaYf6k39SOr1ogGYmF90EjRxy1i21k9wQQjPf26bcBu/9X67KrQjQV0uB38CaNukgiSeoLjfptN811u+PJHx6BP+jx3Qa6/3VenNPxHC8WEU0GXql8QSjIHEyCwKb6fMASXOK94JyB5Ywov2x8mt/+9ncqBBBMVzf6r5Sagy4PL1XnryLsuADD/vOEkPet8wXgH/Oj7v5tTsQQZ7U1JT51PoDs2BFnXc5v3TkVgZwd32k3ONh+nkDw1Hof+4zwUGOyJE6eMrlYzRlKM4Qxdf9JpavQvqfieAbTRWcyKeclnHeoIfE7cDBQIDAQAB",
   "background": {

package/dist/{engine-cUomEg92.js → engine-BWoDQ-3C.js}

@@ -1,4 +1,4 @@
-import { B as withNoOutputRetry, F as IMPLEMENT_DEFAULT_MODEL, I as PLAN_DEFAULT_MODEL, L as REVIEW_DEFAULT_MODEL, M as BROWSE_DEFAULT_MODEL, N as DEFAULT_MODEL, P as EXPLORE_DEFAULT_MODEL, R as appendPlanReminder, z as runWorkerAgent } from "./peer-mcp-personas-BCGYWok0.js";
+import { B as withNoOutputRetry, F as IMPLEMENT_DEFAULT_MODEL, I as PLAN_DEFAULT_MODEL, L as REVIEW_DEFAULT_MODEL, M as BROWSE_DEFAULT_MODEL, N as DEFAULT_MODEL, P as EXPLORE_DEFAULT_MODEL, R as appendPlanReminder, z as runWorkerAgent } from "./peer-mcp-personas-BKkdfOyK.js";
 import "./paths-CTlT1nTo.js";
 import "./lifecycle-ChPBRt6K.js";
 import "./lifecycle-BUXxiltc.js";

package/dist/main.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { $ as isAdvisorRequested, $t as cacheModels, A as liveExec, At as hasSupportedBrowserInstalled, Bt as DEFAULT_CODEX_MODEL_FALLBACKS, C as fileReviewDebounce, Ct as pickEndpoint, D as stopGateEnabledForRepo, Dt as readResponseBodyCapped, E as repoRoot, Et as MAX_RESPONSE_BODY_BYTES, Ft as ArtifactClient, G as vscodeRipgrepPath, Gt as pickClaudeDefault, H as buildToolbeltAwareness, Ht as UPSTREAM_FETCH_TIMEOUT_MS, It as collapsePathKeys, J as searchWeb, Jt as setupCopilotToken, K as TOOLBELT_TOOLS, Kt as getPackageVersion, Lt as toolbeltPathOverride, Mt as extractTarGzMember, Nt as extractZipMember, O as stopReviewStateDir, Ot as parseJsonOrDiagnose, Pt as shouldUseInsecureTls, Q as injectAdvisorTool, Qt as cacheCopilotVersion, Rt as DEFAULT_CLAUDE_MODEL_FALLBACKS, S as fileLastPromptStore, St as resolveMcpToolTimeoutMs, T as repoFingerprint, Tt as createChatCompletions, U as toolbeltEnabled, Ut as UPSTREAM_INACTIVITY_TIMEOUT_MS, V as availableToolCommands, Vt as DEFAULT_PORT, W as toolbeltSkipSet, Wt as generateRandomPort, X as ADVISOR_TOOL_INSTRUCTIONS, Xt as setupGitHubToken, Y as ADVISOR_INTERNAL_TOOL_NAME, Yt as setupGitHubAgentToken, Z as buildAdvisorStream, Zt as tryRefreshAndRetry, _ as stopGateId, _t as workerToolsEnabled, a as buildPeerAwarenessSnippet, an as sleep, at as relayAnthropicStream, b as fileBaselineStore, bt as getTokenCount, c as buildArtifactOpenHookCommand, cn as HTTPError, ct as agentToolsEnabled, d as captureLaunchBaseline, dn as copilotBaseUrl, dt as browserCompoundToolsEnabled, en as cacheVSCodeVersion, et as buildAnthropicErrorEvent, f as decideStopHook, fn as copilotHeaders, ft as browserToolsEnabled, g as stopGateDisabled, gt as standInToolEnabled, h as launchBaselineKey, ht as implementerSubagentModel, i as buildAgentPrompt, in as resolveModel, it as readIteratorWithTimeout, j as resolveSealedGate, jt as provisionAndIndexColbert, k as trustRepo, kt as provisionBrowserAssets, l as buildSessionBindHookCommand, ln as forwardError, lt as artifactToolsEnabled, m as injectStopHookIntoSettingsFile, mn as state, mt as geminiAvailable, n as MCP_GROUPS, nn as isNullish, nt as isControllerClosedError, o as buildPeerAwarenessSummary, on as getModels, ot as handleMcpDelete, p as fileBlockBudget, pn as githubHeaders, pt as fleetToolsEnabled, q as assetFor, qt as withInstallLock, r as assertMcpToolSurfaceConsistent, rn as resolveCodexModel, rt as logStreamError, s as personasFor, sn as fetchWithTransientRetry, st as handleMcpPost, t as GROUP_META, tn as filterBetaHeader, tt as buildOpenAIErrorEvent, u as buildStopHookCommand, un as GITHUB_API_BASE_URL, ut as browseAgentEnabled, v as stopGatePlanMode, vt as countTokens, w as isSubagentContext, wt as createResponses, x as fileFindingsStore, xt as assembleResponsesPayload, y as stopReviewEnabled, yt as createMessages, zt as DEFAULT_CODEX_MODEL } from "./peer-mcp-personas-BCGYWok0.js";
+import { $ as isAdvisorRequested, $t as cacheModels, A as liveExec, At as hasSupportedBrowserInstalled, Bt as DEFAULT_CODEX_MODEL_FALLBACKS, C as fileReviewDebounce, Ct as pickEndpoint, D as stopGateEnabledForRepo, Dt as readResponseBodyCapped, E as repoRoot, Et as MAX_RESPONSE_BODY_BYTES, Ft as ArtifactClient, G as vscodeRipgrepPath, Gt as pickClaudeDefault, H as buildToolbeltAwareness, Ht as UPSTREAM_FETCH_TIMEOUT_MS, It as collapsePathKeys, J as searchWeb, Jt as setupCopilotToken, K as TOOLBELT_TOOLS, Kt as getPackageVersion, Lt as toolbeltPathOverride, Mt as extractTarGzMember, Nt as extractZipMember, O as stopReviewStateDir, Ot as parseJsonOrDiagnose, Pt as shouldUseInsecureTls, Q as injectAdvisorTool, Qt as cacheCopilotVersion, Rt as DEFAULT_CLAUDE_MODEL_FALLBACKS, S as fileLastPromptStore, St as resolveMcpToolTimeoutMs, T as repoFingerprint, Tt as createChatCompletions, U as toolbeltEnabled, Ut as UPSTREAM_INACTIVITY_TIMEOUT_MS, V as availableToolCommands, Vt as DEFAULT_PORT, W as toolbeltSkipSet, Wt as generateRandomPort, X as ADVISOR_TOOL_INSTRUCTIONS, Xt as setupGitHubToken, Y as ADVISOR_INTERNAL_TOOL_NAME, Yt as setupGitHubAgentToken, Z as buildAdvisorStream, Zt as tryRefreshAndRetry, _ as stopGateId, _t as workerToolsEnabled, a as buildPeerAwarenessSnippet, an as sleep, at as relayAnthropicStream, b as fileBaselineStore, bt as getTokenCount, c as buildArtifactOpenHookCommand, cn as HTTPError, ct as agentToolsEnabled, d as captureLaunchBaseline, dn as copilotBaseUrl, dt as browserCompoundToolsEnabled, en as cacheVSCodeVersion, et as buildAnthropicErrorEvent, f as decideStopHook, fn as copilotHeaders, ft as browserToolsEnabled, g as stopGateDisabled, gt as standInToolEnabled, h as launchBaselineKey, ht as implementerSubagentModel, i as buildAgentPrompt, in as resolveModel, it as readIteratorWithTimeout, j as resolveSealedGate, jt as provisionAndIndexColbert, k as trustRepo, kt as provisionBrowserAssets, l as buildSessionBindHookCommand, ln as forwardError, lt as artifactToolsEnabled, m as injectStopHookIntoSettingsFile, mn as state, mt as geminiAvailable, n as MCP_GROUPS, nn as isNullish, nt as isControllerClosedError, o as buildPeerAwarenessSummary, on as getModels, ot as handleMcpDelete, p as fileBlockBudget, pn as githubHeaders, pt as fleetToolsEnabled, q as assetFor, qt as withInstallLock, r as assertMcpToolSurfaceConsistent, rn as resolveCodexModel, rt as logStreamError, s as personasFor, sn as fetchWithTransientRetry, st as handleMcpPost, t as GROUP_META, tn as filterBetaHeader, tt as buildOpenAIErrorEvent, u as buildStopHookCommand, un as GITHUB_API_BASE_URL, ut as browseAgentEnabled, v as stopGatePlanMode, vt as countTokens, w as isSubagentContext, wt as createResponses, x as fileFindingsStore, xt as assembleResponsesPayload, y as stopReviewEnabled, yt as createMessages, zt as DEFAULT_CODEX_MODEL } from "./peer-mcp-personas-BKkdfOyK.js";
 import { a as removeOwnClaudeConfigMirror, i as isUnderClaudeConfigMirror, l as writeArtifactCredsToMirror, n as ensureClaudeConfigMirror, r as ensurePaths, t as PATHS, u as writeRuntimeFileSecure } from "./paths-CTlT1nTo.js";
 import { c as killManagedTree, d as runCommandCapture, f as runCommandVoid, l as parseBoolEnv, s as killChildProcessTree, u as resolveExecutable } from "./lifecycle-ChPBRt6K.js";
 import { a as sweepRegistry } from "./lifecycle-BUXxiltc.js";
@@ -1137,16 +1137,16 @@ function buildCoordinatorAgent(opts) {
 			"",
 			"The lead's brief will include an artifact (plan, design, diff, or code) and a goal (e.g. 'review before exit-plan', 'review the commit I just made', 'cross-check codex-critic's verdict'). Pick the right peers for the artifact type:",
 			"",
-			"- **Plan / design / architecture choice** → fan out to `codex-critic` (gpt-5.5, strongest reasoning, cross-lab)" + (opts.geminiAvailable ? " AND `gemini-critic` (third-lab triangulation, strong on formal reasoning) in parallel" : "") + ". codex-reviewer is the wrong tool for plans (it's a code-specialist, not an architecture critic).",
+			"- **Plan / design / architecture choice** → fan out to `codex-critic` (gpt-5.6-sol, strongest reasoning, cross-lab)" + (opts.geminiAvailable ? " AND `gemini-critic` (third-lab triangulation, strong on formal reasoning) in parallel" : "") + ". codex-reviewer is the wrong tool for plans (it's a code-specialist, not an architecture critic).",
 			"- **Concrete diff or single file** → fan out to `codex-reviewer` (gpt-5.3-codex, line-level code specialist, fastest at ~16s)" + (opts.geminiAvailable ? " AND `gemini-reviewer` (gemini-3.1-pro, second-lab line-level review)" : "") + (opts.geminiAvailable ? " AND `gemini-critic` for cross-lab triangulation" : "") + ". For very small changes (<20 lines), one `codex-reviewer` call is enough.",
-			"- **Large artifact** → the only peers that take a large artifact WHOLE are `codex-critic` (gpt-5.5, ≈922K-token input window) and `opus-critic` (Opus-4.7-1M, ≈936K-token input on enterprise catalogs; ≈168K otherwise). Route the full artifact to those for cross-lab coverage. `codex-reviewer` (≈272K) and `gemini-critic` (≈136K) have small windows — see Decomposition below: never summarize or downsize the request to squeeze a large artifact into a small-window peer.",
+			"- **Large artifact** → the only peers that take a large artifact WHOLE are `codex-critic` (gpt-5.6-sol, ≈1M-token input window) and `opus-critic` (Opus-4.7-1M, ≈936K-token input on enterprise catalogs; ≈168K otherwise). Route the full artifact to those for cross-lab coverage. `codex-reviewer` (≈272K) and `gemini-critic` (≈136K) have small windows — see Decomposition below: never summarize or downsize the request to squeeze a large artifact into a small-window peer.",
 			"- **Formal reasoning, proofs, or invariants** → prefer `gemini-critic`" + (opts.geminiAvailable ? " (gemini-3.1-pro, strong on math and formally-stated properties)" : " (NOT REGISTERED in this session — gemini-3.x not in catalog)") + ".",
 			"- **Tie-breaker after codex-critic has weighed in** → call `gemini-critic`" + (opts.geminiAvailable ? "" : " (NOT REGISTERED in this session)") + " or `opus-critic` with the artifact AND codex-critic's verdict for cross-check.",
 			"- **Fast sanity check** → `opus-critic` (~22s, same lab as lead but fresh context — catches confabulation and motivated reasoning).",
 			"",
 			"## Decomposition for large artifacts",
 			"",
-			"Route by the peer's real PROMPT WINDOW (input tokens): `codex-critic` gpt-5.5 ≈922K · `opus-critic` Opus-4.7-1M ≈936K (enterprise catalogs; ≈168K otherwise) · `codex-reviewer` gpt-5.3-codex ≈272K · `gemini-critic` gemini-3.1-pro ≈136K. The proxy REJECTS (with an actionable message) any single call whose brief exceeds the target peer's window — it will NOT silently truncate, because dropping lines from a review artifact is worse than a clear error. So: send the full artifact only to peers whose window fits it (large artifacts → `codex-critic` and/or `opus-critic`). When a peer's window is too small (commonly `gemini-critic` at ≈136K, or `codex-reviewer` at ≈272K), do NOT summarize or downsize the request to include it — either skip that peer, or split the artifact into 2-4 logical batches BY CONCERN (not by raw size — semantic batches give better per-batch reviews) that each fit, and call in parallel. Use the big-window peers for the whole and reserve a small-window peer like gemini for the concerns it can actually hold. The proxy's MCP cap allows up to 8 in-flight calls. Aggregate findings yourself before reporting back. (Separately, on the JSON transport a per-effort `predictedTooLong` byte cap still guards the ~60s tools/call timeout for non-SSE clients; Claude Code uses SSE, which streams with heartbeats and isn't subject to that cap.)",
+			"Route by the peer's real PROMPT WINDOW (input tokens): `codex-critic` gpt-5.6-sol ≈1M · `opus-critic` Opus-4.7-1M ≈936K (enterprise catalogs; ≈168K otherwise) · `codex-reviewer` gpt-5.3-codex ≈272K · `gemini-critic` gemini-3.1-pro ≈136K. The proxy REJECTS (with an actionable message) any single call whose brief exceeds the target peer's window — it will NOT silently truncate, because dropping lines from a review artifact is worse than a clear error. So: send the full artifact only to peers whose window fits it (large artifacts → `codex-critic` and/or `opus-critic`). When a peer's window is too small (commonly `gemini-critic` at ≈136K, or `codex-reviewer` at ≈272K), do NOT summarize or downsize the request to include it — either skip that peer, or split the artifact into 2-4 logical batches BY CONCERN (not by raw size — semantic batches give better per-batch reviews) that each fit, and call in parallel. Use the big-window peers for the whole and reserve a small-window peer like gemini for the concerns it can actually hold. The proxy's MCP cap allows up to 8 in-flight calls. Aggregate findings yourself before reporting back. (Separately, on the JSON transport a per-effort `predictedTooLong` byte cap still guards the ~60s tools/call timeout for non-SSE clients; Claude Code uses SSE, which streams with heartbeats and isn't subject to that cap.)",
 			"",
 			"## Aggregation contract",
 			"",
@@ -1206,7 +1206,7 @@ function buildPeerAgentDefinitions(opts) {
 		geminiAvailable: opts.geminiAvailable
 	});
 	if (opts.implementerModel && opts.implementerModel.length > 0) out.implementer = {
-		description: "Bounded implementation subagent running gpt-5.5 (strong non-Claude coder, high reasoning). Use for well-scoped coding tasks — edits, small features, fixes — you want implemented in an integrated subagent. Model is overridable at spawn.",
+		description: `Bounded implementation subagent running ${opts.implementerModel} (strong non-Claude coder, high reasoning). Use for well-scoped coding tasks — edits, small features, fixes — you want implemented in an integrated subagent. Model is overridable at spawn.`,
 		prompt: "You are a bounded implementation subagent for well-scoped coding tasks. Implement the requested change surgically, matching the surrounding code style and minimizing unrelated churn. Use the dedicated Edit/Write/Read tools for file changes and Grep/Glob for search; reserve Bash for running builds, tests, and git — do not shell out (sed/awk/python/here-docs) to read or edit files. Verify with the project's build or tests where applicable. Do the work yourself — do not spawn further subagents. Report exactly what changed and any risks.",
 		model: opts.implementerModel
 	};
@@ -1787,7 +1787,7 @@ async function callMcpTool(opts) {
 	};
 }
 /**
-* One non-streaming gpt-5.5 (or any model id) inference via `/v1/responses`.
+* One non-streaming gpt-5.6-sol (or any model id) inference via `/v1/responses`.
 * Returns the assistant text (possibly empty). Throws on transport/HTTP/parse
 * failure. `effort` maps to the Responses `reasoning.effort` knob.
 */
@@ -2556,7 +2556,7 @@ async function discoverGateCommands(cwd, opts) {
 	if (files.length === 0) return null;
 	let result;
 	try {
-		const { runWorkerAgent } = await import("./engine-cUomEg92.js");
+		const { runWorkerAgent } = await import("./engine-BWoDQ-3C.js");
 		result = await runWorkerAgent({
 			mode: "explore",
 			workspace: root,
@@ -2622,14 +2622,14 @@ function decidePromptSubmit(input) {
 * concluding. Mirrors the v1 advisory tone: additive, never blocking.
 */
 const PROMPT_SEARCH_TIP = "TIP (advisory): when this task needs code context, search lexical + semantic in parallel — one `mcp__search__code` call with mode:\"lexical\" and one with mode:\"semantic\", issued in the same turn — before concluding.";
-/** System prompt for the single gpt-5.5 scope/goal inference. Steers a SHORT,
+/** System prompt for the single gpt-5.6-sol scope/goal inference. Steers a SHORT,
 *  user-derived (not invented) advisory note grounded in the search results. */
 const PROMPT_SCOPE_SYSTEM = "You are a scoping assistant for a coding agent about to act on a user's request. You are given the user's request and the results of a lexical + semantic code search over the relevant repository. Produce a SHORT advisory note (<= 120 words), plain text only:\n1. SCOPE: one line — is this trivial, focused (one area), or large/cross-cutting — grounded in what the search surfaced (reference the most relevant file(s) by name).\n2. GOAL: restate the user's OWN ask as a single measurable objective, in THEIR terms. Do NOT invent new requirements or acceptance criteria beyond what they asked.\n3. Only if the task is large/cross-cutting, add a final line: \"Consider /gh-research first to saturate understanding, then /gh-orchestrate to compose a floor-raising workflow.\" Omit it for a focused or trivial task.\nThis is advisory — the agent decides whether to follow it. Be concrete and concise; no preamble.";
 /** Max chars of each search-result blob fed into the scope inference. */
 const SEARCH_CONTEXT_CAP = 6 * 1024;
 /** Wrap the prior-turn review findings in an explicitly NON-AUTHORITATIVE frame. */
 function framePendingFindings(findings) {
-	return "ADVISORY — independent review of your PREVIOUS change (NON-AUTHORITATIVE): an independent gpt-5.5 reviewer flagged the following. Evaluate each on its merits — fix the real ones, and ignore any wrong one with a one-line reason. You are NOT obligated to act on these.\n" + findings.trim();
+	return "ADVISORY — independent review of your PREVIOUS change (NON-AUTHORITATIVE): an independent gpt-5.6-sol reviewer flagged the following. Evaluate each on its merits — fix the real ones, and ignore any wrong one with a one-line reason. You are NOT obligated to act on these.\n" + findings.trim();
 }
 function joinSections(sections) {
 	return sections.map((s) => s.trim()).filter((s) => s.length > 0).join("\n\n");
@@ -2643,7 +2643,7 @@ function joinSections(sections) {
 *   - subagent/teammate  -> empty (top-level only, like v1).
 *   - findings           -> always surfaced (+ cleared) regardless of triviality.
 *   - trivial prompt     -> findings only (no search tip, no model call).
-*   - substantive prompt -> static search tip + parallel lexical+semantic search -> ONE gpt-5.5 call
+*   - substantive prompt -> static search tip + parallel lexical+semantic search -> ONE gpt-5.6-sol call
 *                           -> grounded scope/goal note. Fail-open to PROMPT_STEER_GOAL.
 *   - steerEnabled=false -> findings only (no goal/tip).
 */
@@ -4716,7 +4716,7 @@ function initProxyFromEnv() {
 //#endregion
 //#region package.json
 var name = "github-router";
-var version$1 = "0.3.175";
+var version$1 = "0.3.176";
 
 //#endregion
 //#region src/lib/approval.ts
@@ -8062,6 +8062,10 @@ function parseSharedArgs(args) {
 * 1M/400k window, never overflows). See `seedGatewayModelCache`.
 */
 const NATIVE_NON_CLAUDE_MODELS = [
+	{
+		id: "gpt-5.6-sol",
+		displayName: "GPT-5.6 Sol"
+	},
 	{
 		id: "gpt-5.5",
 		displayName: "GPT-5.5"
@@ -8784,13 +8788,10 @@ const codex = defineCommand({
 		if (codexModel !== requestedModel) consola.info(`Model "${requestedModel}" resolved to "${codexModel}"`);
 		if (usingDefault && state.models) {
 			const inCache = (id) => state.models?.data.some((m) => m.id === id) ?? false;
-			if (!inCache(codexModel)) for (const fallback of DEFAULT_CODEX_MODEL_FALLBACKS) {
-				const resolved = resolveCodexModel(fallback);
-				if (inCache(resolved)) {
-					consola.info(`Default model "${codexModel}" not in your Copilot model list; falling back to "${resolved}".`);
-					codexModel = resolved;
-					break;
-				}
+			const firstPresent = [DEFAULT_CODEX_MODEL, ...DEFAULT_CODEX_MODEL_FALLBACKS].map((id) => resolveModel(id)).find((id) => inCache(id));
+			if (firstPresent) {
+				if (firstPresent !== codexModel) consola.info(`Default model "${DEFAULT_CODEX_MODEL}" not in your Copilot model list; falling back to "${firstPresent}".`);
+				codexModel = firstPresent;
 			}
 		}
 		const modelEntry = state.models?.data.find((m) => m.id === codexModel);
@@ -8955,7 +8956,7 @@ const internalPromptSubmit = defineCommand({
 						},
 						infer: (system, user, signal) => callInference({
 							serverUrl: runtime.serverUrl,
-							model: "gpt-5.5",
+							model: "gpt-5.6-sol",
 							instructions: system,
 							input: user,
 							effort: "low",
@@ -9688,7 +9689,7 @@ THE DIFF:
 const internalStopReview = defineCommand({
 	meta: {
 		name: "internal-stop-review",
-		description: "Internal: the detached, advisory background reviewer. Reads a JSON payload on stdin, runs a read-only gpt-5.5 review of the working tree against the user's ask, and writes advisory findings for the next prompt to surface. Never blocks anything."
+		description: "Internal: the detached, advisory background reviewer. Reads a JSON payload on stdin, runs a read-only gpt-5.6-sol review of the working tree against the user's ask, and writes advisory findings for the next prompt to surface. Never blocks anything."
 	},
 	async run() {
 		try {
@@ -9717,7 +9718,7 @@ const internalStopReview = defineCommand({
 						transcriptPath: typeof payload.transcript_path === "string" ? payload.transcript_path : ""
 					}),
 					workspace: cwd,
-					model: "gpt-5.5",
+					model: "gpt-5.6-sol",
 					thinking: "high"
 				},
 				timeoutMs: REVIEW_TIMEOUT_MS