github-router 0.3.66 → 0.3.68

package/README.md

@@ -34,10 +34,10 @@ The server runs at `http://localhost:8787`. Now pick your tool below.
 npx github-router@latest claude
 ```
 
-Boots the proxy on a random port and spawns Claude Code wired to it. Sets `ANTHROPIC_MODEL=claude-opus-4-7` (Anthropic's dashed slug — Claude Code's `/model` UI displays this as menu entry "Opus 4.7 (1M context)"). The proxy translates to Copilot's `claude-opus-4.7-1m-internal` on enterprise tokens or `claude-opus-4.7` on Pro+/Business/Max at request time. Major.minor fallback chain: `claude-opus-4-6` → `claude-opus-4-5`. Override with `-m`:
+Boots the proxy on a random port and spawns Claude Code wired to it. Sets `ANTHROPIC_MODEL=claude-opus-4-8` (Anthropic's dashed slug — Claude Code's `/model` UI displays this as menu entry "Opus 4.8 (1M context)"). The proxy translates to Copilot's `claude-opus-4.8` at request time (single base slug; the catalog entry's `max_context_window_tokens` already advertises 1M context — no `-1m` sibling needed for 4.8). Major.minor fallback chain: `claude-opus-4-7` → `claude-opus-4-6` → `claude-opus-4-5`. Override with `-m`:
 
 ```sh
-npx github-router@latest claude -m claude-opus-4-7
+npx github-router@latest claude -m claude-opus-4-8
 ```
 
 The launcher sanitizes parent-env auth keys and sets `CLAUDE_CONFIG_DIR=$HOME/.claude` so the spawned `claude` ignores any persisted Console OAuth credential without requiring `claude /logout`. Settings, MCP servers, hooks, and CLAUDE.md auto-discovery still load from `~/.claude` as normal.
@@ -85,7 +85,7 @@ Each persona is exposed both as a Claude Code subagent (callable via the `Task`
 |---|---|---|---|
 | `codex-critic` | gpt-5.5 | `/v1/responses` | low \| medium \| high \| xhigh (xhigh) |
 | `codex-reviewer` | gpt-5.3-codex | `/v1/responses` | low \| medium \| high \| xhigh (xhigh) |
-| `opus-critic` | claude-opus-4-7 | `/v1/messages` | low \| medium \| high \| xhigh (xhigh) |
+| `opus-critic` | claude-opus-4-6 | `/v1/messages` | low \| medium \| high (high) |
 | `gemini-critic` | gemini-3.1-pro-preview | `/v1/chat/completions` | low \| medium \| high (high) |
 | `peer-review-coordinator` | (meta) | — | — |
 
@@ -169,6 +169,7 @@ Anthropic endpoints are only available under `/v1/messages`.
 | gpt-4.1, gpt-4o | Yes | Yes | No |
 | 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 |
 | claude-opus-4.6-1m, claude-opus-4.6, claude-sonnet-4.6 | Yes | No | Yes |
 | o3, o4-mini | Yes | Yes | No |
@@ -234,7 +235,7 @@ 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-4-7` (Anthropic dashed slug for UI compatibility; the proxy translates to Copilot's `claude-opus-4.7-1m-internal` on enterprise or `claude-opus-4.7` elsewhere). Major.minor fallback chain: `claude-opus-4-6` → `claude-opus-4-5`.
+- `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`
 
 Fallback chains fire only on the implicit-default path; explicit `-m`/`--model` is always respected as-is.
@@ -257,7 +258,7 @@ Additional flags accepted only by the `claude` subcommand:
 
 | Flag | Description | Default |
 |---|---|---|
-| `--model, -m` | Override the default Claude model | claude-opus-4-7 |
+| `--model, -m` | Override the default Claude model | claude-opus-4-8 |
 | `--codex-mcp` / `--no-codex-mcp` | Wire peer-MCP review subagents (codex-critic / opus-critic / gemini-critic / codex-reviewer / peer-review-coordinator) into the spawned session | true |
 | `--codex-cli` | Add a `codex mcp-server` stdio backend so `codex-implementer` can mutate files. Requires codex CLI 0.129+; falls back to HTTP-only if absent | false |
 | `--codex-mcp-only` | Pass `--strict-mcp-config` to Claude Code so only the proxy's MCP servers load (hides any user MCP servers in `~/.claude/mcp.json`) | false |

package/dist/browser-ext/manifest.json

@@ -2,7 +2,7 @@
   "manifest_version": 3,
   "name": "github-router browser bridge",
   "short_name": "gh-router-browser",
-  "version": "0.3.66",
+  "version": "0.3.68",
   "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/main.js

@@ -931,15 +931,34 @@ async function autoUpdateClaude(latestVersion) {
 //#endregion
 //#region src/lib/port.ts
 const DEFAULT_PORT = 8787;
-const DEFAULT_CLAUDE_MODEL_FALLBACKS = ["claude-opus-4-6", "claude-opus-4-5"];
+const DEFAULT_CLAUDE_MODEL_FALLBACKS = [
+	"claude-opus-4-7",
+	"claude-opus-4-6",
+	"claude-opus-4-5"
+];
 /**
 * Cap-aware default picker for `ANTHROPIC_MODEL` on the implicit-default
 * path. Returns `claude-opus-${family}[1m]` when the live Copilot catalog
-* contains an `opus-${family}-1m*` variant (enterprise tier), else the
-* bare `claude-opus-${family}` slug. `family` defaults to `"4.7"` so the
-* no-arg call preserves the original behavior; explicit values like
-* `"4.6"` or `"4.8"` are used to honor the `github-router claude
-* -m <version>` family shorthand.
+* shows the family is 1M-capable, else the bare `claude-opus-${family}`
+* slug. `family` defaults to `"4.8"` so the no-arg call selects the
+* current default; explicit values like `"4.7"` or `"4.6"` are used to
+* honor the `github-router claude -m <version>` family shorthand.
+*
+* **Dual-signal 1M detection**. The Opus families have evolved different
+* shapes in Copilot's catalog over time:
+*   1. **Sibling-slug signal** — `opus-${family}-1m` (or `opus-${family}-1m-internal`)
+*      exists as a separate catalog entry distinct from the base slug.
+*      This is how 4.6 and 4.7 ship (`claude-opus-4.6-1m`,
+*      `claude-opus-4.7-1m-internal`). Matched by the version-anchored
+*      regex below.
+*   2. **Base-slug capability signal** — the catalog entry whose id IS
+*      the base `opus-${family}` slug advertises
+*      `capabilities.limits.max_context_window_tokens >= 1_000_000`. This
+*      is how 4.8 ships — there is no `-1m` sibling; the single
+*      `claude-opus-4.8` id is the 1M variant.
+* Either signal flips on the `[1m]` decoration. Both signals together
+* also flip it on (no double-counting). The breadcrumb log names which
+* signal fired so users can spot catalog shape changes.
 *
 * The `[1m]` literal-bracket suffix is Claude Code's local 1M-context
 * unlock — cc-backup `src/utils/context.ts:35-40` matches `/\[1m\]/i`
@@ -949,14 +968,14 @@ const DEFAULT_CLAUDE_MODEL_FALLBACKS = ["claude-opus-4-6", "claude-opus-4-5"];
 * proxy routes the underlying request.
 *
 * Cap-awareness matters because on non-enterprise Copilot tiers there
-* is no `-1m` opus backend; sending `[1m]` there would either 400 at
+* is no 1M opus backend; sending `[1m]` there would either 400 at
 * Copilot or (with `resolveModel`'s graceful-degrade) silently
 * downgrade upstream while Claude Code still over-accounts context.
 * This helper detects the catalog state at launch and only opts in
 * when the backend can actually serve 1M.
 *
 * Sonnet/Haiku families are intentionally NOT given `[1m]` defaults
-* because Copilot has no `-1m` backend for them (and Anthropic-side
+* because Copilot has no 1M backend for them (and Anthropic-side
 * `modelSupports1M` doesn't list haiku at all). See
 * `src/lib/server-setup.ts:getClaudeCodeEnvVars` for the
 * `ANTHROPIC_DEFAULT_{SONNET,HAIKU,OPUS}_MODEL` tier defaults.
@@ -966,18 +985,25 @@ const DEFAULT_CLAUDE_MODEL_FALLBACKS = ["claude-opus-4-6", "claude-opus-4-5"];
 * can't tell the difference between "no catalog yet" and "no 1M
 * variant" — defaulting safe-side preserves the pre-change behavior).
 */
-const DEFAULT_OPUS_FAMILY = "4.7";
+const DEFAULT_OPUS_FAMILY = "4.8";
+const ONE_M_TOKENS = 1e6;
 function pickClaudeDefault(opusFamily = DEFAULT_OPUS_FAMILY) {
 	const dotted = opusFamily.replace(/-/g, ".");
 	const bareSlug = `claude-opus-${dotted.replace(/\./g, "-")}`;
 	const versionPattern = dotted.replace(/\./g, "[.-]");
 	const oneMRegex = new RegExp(`opus-${versionPattern}-1m(?:$|-)`, "i");
+	const baseSlugRegex = new RegExp(`^claude-opus-${versionPattern}$`, "i");
 	const familyRegex = new RegExp(`opus-${versionPattern}(?:$|[-.])`, "i");
 	const models$1 = state.models?.data ?? [];
-	const has1m = models$1.some((m) => oneMRegex.test(m.id));
+	const siblingOneM = models$1.some((m) => oneMRegex.test(m.id));
+	const baseSlugMaxContext = models$1.reduce((max, m) => baseSlugRegex.test(m.id) ? Math.max(max, m.capabilities?.limits?.max_context_window_tokens ?? 0) : max, 0);
+	const baseSlugOneM = baseSlugMaxContext >= ONE_M_TOKENS;
+	const has1m = siblingOneM || baseSlugOneM;
 	if (opusFamily !== DEFAULT_OPUS_FAMILY && state.models && models$1.length > 0 && !models$1.some((m) => familyRegex.test(m.id))) consola.warn(`Requested Opus family "${dotted}" not found in Copilot catalog; using "${bareSlug}" anyway (resolveModel may not find a backend for it).`);
 	if (has1m) {
-		consola.info(`Catalog contains opus-${dotted}-1m variant; defaulting ANTHROPIC_MODEL to "${bareSlug}[1m]" so Claude Code accounts for 1M context locally. Set CLAUDE_CODE_DISABLE_1M_CONTEXT=1 to opt out (HIPAA), or pass --model ${bareSlug} to pin 200K.`);
+		const signal = siblingOneM ? baseSlugOneM ? "sibling-slug + base-slug 1M capability" : `sibling slug opus-${dotted}-1m` : `base slug ${bareSlug} (max_context_window_tokens=${baseSlugMaxContext})`;
+		const pinHint = siblingOneM ? ` Pass --model ${bareSlug} to pin 200K.` : ` (No separate 200K variant of ${dotted} exists in the catalog — the bare slug IS the 1M backend.)`;
+		consola.info(`Catalog signals opus-${dotted} is 1M-capable (${signal}); defaulting ANTHROPIC_MODEL to "${bareSlug}[1m]" so Claude Code accounts for 1M context locally. Set CLAUDE_CODE_DISABLE_1M_CONTEXT=1 to opt out (HIPAA).${pinHint}`);
 		return `${bareSlug}[1m]`;
 	}
 	return bareSlug;
@@ -8494,17 +8520,21 @@ function browserToolsEnabled() {
 	return hasSupportedBrowserInstalled();
 }
 /**
-* The 1M-context Opus variant (`claude-opus-4.7-1m-internal`,
-* `max_prompt_tokens` 936K), gated `restricted_to: ["enterprise"]`.
-* opus_critic prefers it so it can take large artifacts in one shot
+* The 1M-context Opus 4.6 variant (`claude-opus-4.6-1m`, `max_prompt_tokens`
+* 936K). opus_critic prefers it so it can take large artifacts in one shot
 * (the whole point of pairing it with gpt-5.5 as the big-window peers);
-* falls back to the 200K `claude-opus-4-7` when the catalog (non-
-* enterprise) doesn't carry a 1M opus slug.
-*/
-const OPUS_1M_RE = /opus-4\.7.*1m/i;
+* falls back to the 200K `claude-opus-4-6` when the catalog doesn't carry
+* a 1M 4.6 slug. The regex is version-anchored to 4.6 AND requires a
+* `-1m` suffix boundary (not a permissive `.*1m`), so it does NOT
+* false-positive on `claude-opus-4.7-1m-internal` (stand_in's pinned
+* 4.7 row), `claude-opus-4.6-1max` (hypothetical), or `claude-opus-4.8`
+* (1M-without-sibling). Tolerates dotted (`opus-4.6-1m`) and dashed
+* (`opus-4-6-1m`) catalog separators.
+*/
+const OPUS_1M_RE = /opus-4[.-]6-1m(?:$|-)/i;
 function resolveOpusCriticModel() {
 	const oneM = state.models?.data?.find((m) => OPUS_1M_RE.test(m.id));
-	return oneM ? oneM.id : "claude-opus-4-7";
+	return oneM ? oneM.id : "claude-opus-4-6";
 }
 function activePersonas() {
 	return PERSONAS_READ.filter((p) => !p.requiresGeminiCatalog || geminiAvailable()).map((p) => p.toolNameHttp === "opus_critic" ? {
@@ -8600,6 +8630,8 @@ function toolError(message) {
 *   gpt-5.3-codex high on ~600B = 16.0s → ~64s on 12KB
 *   claude-opus-4-7 medium (thinking=3000) on a trivial prompt = 22.5s
 *     but model self-paces budget → ~50s+ on a real ~6KB review
+*     (still applicable to stand_in's 4.7 row; opus_critic now runs on
+*     4.6 with similar empirical shape)
 *
 * Returns `{tooLong: true, capBytes}` when the (persona, effort, briefBytes)
 * tuple is empirically predicted to bust the 60s ceiling.
@@ -12082,10 +12114,11 @@ function round2(n) {
 * **xhigh on long-running personas works via SSE-streamed /mcp responses**
 * (handler.ts:handleToolsCallSSE). Claude Code's MCP HTTP client honors
 * `text/event-stream` responses without applying the ~60s per-tool-call
-* timer that previously broke xhigh on gpt-5.5 (~56s wall) and
-* claude-opus-4-7 (high+ thinking budgets). All four personas now expose
-* all four effort tiers with `high` default; SSE handles the long tail
-* transparently to the user.
+* timer that previously broke xhigh on gpt-5.5 (~56s wall) and on
+* Anthropic Opus families (high+ thinking budgets). opus-critic itself
+* now runs on claude-opus-4-6 which doesn't advertise xhigh, so the
+* SSE long-tail concern there is moot; the SSE machinery still applies
+* to the other personas that do expose xhigh.
 */
 const EFFORT_LEVELS = [
 	"low",
@@ -12263,9 +12296,9 @@ const PERSONAS_READ = Object.freeze([
 	{
 		agentName: "opus-critic",
 		toolNameHttp: "opus_critic",
-		model: "claude-opus-4-7",
+		model: "claude-opus-4-6",
 		endpoint: "/v1/messages",
-		description: "Adversarial second opinion from a fresh-context Opus 4.7 — same lab as the lead, limited blind-spot diversity vs cross-lab critics. On enterprise catalogs that carry Opus-4.7-1M it runs with a ≈936K-token input window and handles large artifacts without decomposition; otherwise ≈168K. Fast (~22s), catches confabulation and motivated reasoning. Pass artifact verbatim.",
+		description: "Adversarial second opinion from a fresh-context Opus 4.6 — same lab as the lead, limited blind-spot diversity vs cross-lab critics. On enterprise catalogs that carry Opus-4.6-1M it runs with a ≈936K-token input window; otherwise ≈168K. Pinned one minor behind the default Opus so the panel spans more of the version curve. Catches confabulation. Pass artifact verbatim.",
 		baseInstructions: OPUS_CRITIC_BASE,
 		agentPrompt: "",
 		writeCapable: false,
@@ -12273,10 +12306,9 @@ const PERSONAS_READ = Object.freeze([
 		allowedEfforts: [
 			"low",
 			"medium",
-			"high",
-			"xhigh"
+			"high"
 		],
-		defaultEffort: "xhigh"
+		defaultEffort: "high"
 	}
 ]);
 const PERSONAS_WRITE = Object.freeze([{
@@ -13855,7 +13887,7 @@ function initProxyFromEnv() {
 //#endregion
 //#region package.json
 var name = "github-router";
-var version$1 = "0.3.66";
+var version$1 = "0.3.68";
 
 //#endregion
 //#region src/lib/approval.ts
@@ -15792,10 +15824,10 @@ function getClaudeCodeEnvVars(serverUrl, model) {
 		DISABLE_TELEMETRY: "1"
 	};
 	if (model) vars.ANTHROPIC_MODEL = model;
-	if (process.env.ANTHROPIC_SMALL_FAST_MODEL === void 0) vars.ANTHROPIC_SMALL_FAST_MODEL = "claude-haiku-4-5";
+	if (process.env.ANTHROPIC_SMALL_FAST_MODEL === void 0) vars.ANTHROPIC_SMALL_FAST_MODEL = "claude-sonnet-4-6";
 	if (process.env.ANTHROPIC_DEFAULT_SONNET_MODEL === void 0) vars.ANTHROPIC_DEFAULT_SONNET_MODEL = "claude-sonnet-4-6";
 	if (process.env.ANTHROPIC_DEFAULT_HAIKU_MODEL === void 0) vars.ANTHROPIC_DEFAULT_HAIKU_MODEL = "claude-haiku-4-5";
-	if (process.env.ANTHROPIC_DEFAULT_OPUS_MODEL === void 0) vars.ANTHROPIC_DEFAULT_OPUS_MODEL = "claude-opus-4-7";
+	if (process.env.ANTHROPIC_DEFAULT_OPUS_MODEL === void 0) vars.ANTHROPIC_DEFAULT_OPUS_MODEL = "claude-opus-4-8";
 	if (process.env.CLAUDE_CODE_PLAN_V2_AGENT_COUNT === void 0) vars.CLAUDE_CODE_PLAN_V2_AGENT_COUNT = "7";
 	for (const key of [
 		"CLAUDE_CODE_ENABLE_EXPERIMENTAL_ADVISOR_TOOL",