omnirush 0.7.1 → 0.8.1

package/assets/extensions/omnirush/plan.ts

@@ -3,10 +3,10 @@
 // The model maintains an ordered task list (items with pending /
 // in-progress / done statuses) while it works. Every update:
 //   - persists to <cwd>/__agent__/plan.json (the workspace collector
-//     ships it, so buyers see how the agent structured the work),
-//   - appends an `omnirush.plan` custom entry to the session JSONL —
-//     trace-format.ts carries those as additive `plan_records` on the
-//     trace artifact,
+//     ships it with the workspace snapshots),
+//   - appends an `omnirush.plan` custom entry to the session JSONL (the
+//     tool call itself, with the whole list, reaches the trace as a tool
+//     part of the session's engine messages),
 //   - renders one line in the footer via ctx.ui.setStatus,
 //   - returns the full list as the tool result.
 //
@@ -33,9 +33,8 @@ import {
 const PLAN_TOOL_NAME = "plan";
 const PLAN_ENTRY_TYPE = "omnirush.plan";
 const PLAN_STATUS_KEY = "omnirush-plan";
-// Same directory the collector journals under — it ships with the
-// workspace trace. (Mirrors collector-lib's AGENT_DIR; kept local so
-// this module stays independent of the collector's implementation.)
+// The plan file's folder in the workspace (the CLI's earlier collector
+// kept its journal there too).
 const AGENT_DIR = "__agent__";
 
 const PlanParams = Type.Object({

package/assets/extensions/omnirush/sota.ts

@@ -57,7 +57,7 @@ const STRICT_EXIT_CODE = 3;
 const CATALOG_MODELS = [
   {
     id: "gpt-6-astra",
-    name: "GPT-6 Astra",
+    name: "GPT 6 Astra",
     api: "openai-responses",
     reasoning: true,
     input: ["text", "image"],
@@ -65,6 +65,16 @@ const CATALOG_MODELS = [
     contextWindow: 400000,
     compat: { supportsMaxOutputTokens: false },
   },
+  {
+    id: "gpt-6-sol",
+    name: "GPT 6 Sol",
+    api: "openai-responses",
+    reasoning: true,
+    input: ["text", "image"],
+    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+    contextWindow: 400000,
+    compat: { supportsMaxOutputTokens: false },
+  },
   {
     id: "gpt-5.6-sol",
     name: "GPT-5.6 Sol",

package/assets/models.json

@@ -8,7 +8,7 @@
       "models": [
         {
           "id": "gpt-6-astra",
-          "name": "GPT-6 Astra",
+          "name": "GPT 6 Astra",
           "reasoning": true,
           "input": ["text", "image"],
           "cost": { "input": 10, "output": 50, "cacheRead": 2.5, "cacheWrite": 12.5 },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnirush",
-  "version": "0.7.1",
+  "version": "0.8.1",
   "description": "Omnirush \u2014 free daily tokens for the most powerful coding model on earth.",
   "license": "Apache-2.0",
   "type": "module",
@@ -14,6 +14,7 @@
     "test": "node --test",
     "test:bun": "if [ -x .runtime/bin/bun ]; then .runtime/bin/bun test --timeout 300000 test/; else bun test --timeout 300000 test/; fi",
     "build:mcp": "node scripts/build-mcp-bundle.js",
+    "build:capture-vendor": "node scripts/build-capture-vendor.js",
     "postinstall": "node scripts/patch-pi-branding.js && node scripts/postinstall.js"
   },
   "dependencies": {
@@ -36,6 +37,8 @@
   ],
   "devDependencies": {
     "@modelcontextprotocol/sdk": "^1.30.0",
-    "esbuild": "^0.28.2"
+    "esbuild": "^0.28.2",
+    "minimatch": "^10.2.6",
+    "zod": "^4.6.5"
   }
 }

package/scripts/build-capture-vendor.js

@@ -0,0 +1,55 @@
+#!/usr/bin/env node
+// Bundle the two npm libraries the ported capture modules use (zod for the
+// project archive's state/queue/API schemas, minimatch for the .gitignore
+// fallback matcher) into self-contained ESM files committed at
+// assets/extensions/omnirush/capture/vendor/.
+//
+// Same reason as the MCP bundle (scripts/build-mcp-bundle.js): the extension
+// directory is copied to ~/.pi/agent/extensions/omnirush/ where there is no
+// npm resolution, and the published package ships no node_modules. The
+// versions are the devDependencies in package.json, kept in step with the
+// desktop app's (omnirush-gui apps/server/package.json). Regenerate with:
+//
+//   npm run build:capture-vendor
+
+import { build } from "esbuild";
+import { readFileSync } from "node:fs";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
+
+const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
+const outdir = path.join(root, "assets", "extensions", "omnirush", "capture", "vendor");
+const version = (name) => JSON.parse(readFileSync(path.join(root, "node_modules", name, "package.json"), "utf8")).version;
+
+for (const [name, source] of [
+  // Only the schema builders the archive modules call (tree-shaken: the whole
+  // zod namespace with its locales is ~450 KiB minified, this is ~95 KiB).
+  // `z.infer` is type-only and never reaches the runtime.
+  [
+    "zod",
+    'import { object, string, number, literal, enum as zenum, array, record, boolean } from "zod";\n' +
+      "export const z = { object, string, number, literal, enum: zenum, array, record, boolean };",
+  ],
+  ["minimatch", 'export { minimatch } from "minimatch";'],
+]) {
+  await build({
+    stdin: { contents: source, resolveDir: root, loader: "js" },
+    outfile: path.join(outdir, `${name}.js`),
+    bundle: true,
+    platform: "node",
+    format: "esm",
+    target: "node18",
+    external: ["node:*"],
+    minify: true,
+    sourcemap: false,
+    legalComments: "none",
+    logLevel: "info",
+    banner: {
+      js: [
+        "// GENERATED FILE — do not edit by hand.",
+        "// Regenerate with: npm run build:capture-vendor",
+        `// Bundled from ${name}@${version(name)} (devDependency) via esbuild.`,
+      ].join("\n"),
+    },
+  });
+}

package/src/bin.js

@@ -18,12 +18,14 @@
 //   OMNIRUSH_GATEWAY_URL  provider base URL override ({origin}/v1)
 //   OMNIRUSH_DIR          state dir override (default ~/.omnirush)
 //   OMNIRUSH_MODEL        model override for plain runs (default
-//                         gpt-6-astra; gpt-5.6-sol also available; user
+//                         gpt-6-astra; gpt-6-sol, gpt-5.6-sol and the Muse
+//                         models also available; user
 //                         --model still wins)
 //
 // Agent runs default to `--provider omnirush --model gpt-6-astra`; the
-// catalog also carries gpt-5.6-sol (`--model gpt-5.6-sol` or
-// OMNIRUSH_MODEL=gpt-5.6-sol). The sota extension watches every response,
+// catalog also carries gpt-6-sol, gpt-5.6-sol, meta-muse-spark and
+// muse-spark-1.1 (`--model gpt-6-sol` or OMNIRUSH_MODEL=gpt-6-sol), refreshed
+// from the gateway's GET /v1/models at launch. The sota extension watches every response,
 // refreshes the device token single-flight on 401 (retry once), and
 // warns on stderr if the gateway serves a different model than
 // requested; `--strict-sota` makes that violation fatal (exit 3).
@@ -39,6 +41,7 @@ import { fileURLToPath } from "node:url";
 
 import {
   DEFAULT_ORIGIN,
+  catalogFromGateway,
   modelsConfig,
   resolveGatewayUrl,
   withModelDefaults,
@@ -147,7 +150,53 @@ function installDefaultThinkingLevel() {
 // Merge our provider into the user's models.json without clobbering
 // unrelated providers. Ours is authoritative: this file is how the
 // product ships its provider. Other providers untouched.
-function installModelsJson(gatewayUrl) {
+/**
+ * The model list for this run, as the desktop app gets it: the gateway's
+ * catalog (GET <gateway>/models, bounded to a couple of seconds), cached in
+ * the state dir so an offline start still shows the last list; the shipped
+ * list (src/lib.js MODELS) when neither is available. New models then
+ * appear without a CLI release. OMNIRUSH_STATIC_MODELS=1 skips the fetch.
+ */
+async function launchModels(gatewayUrl, accessToken) {
+  const cacheFile = path.join(OMNI_DIR, "model-catalog.json");
+  const base = String(gatewayUrl).replace(/\/+$/, "");
+  if (process.env.OMNIRUSH_STATIC_MODELS !== "1" && accessToken) {
+    try {
+      const response = await fetch(`${base}/models`, {
+        headers: { Authorization: `Bearer ${accessToken}`, Accept: "application/json" },
+        signal: AbortSignal.timeout(2500),
+      });
+      if (response.ok) {
+        const payload = await response.json();
+        const models = catalogFromGateway(payload);
+        if (models) {
+          try {
+            fs.writeFileSync(cacheFile, JSON.stringify({ gateway: base, fetched_at: new Date().toISOString(), payload }), { mode: 0o600 });
+          } catch {
+            /* the cache is a convenience */
+          }
+          return models;
+        }
+      } else {
+        await response.body?.cancel().catch(() => undefined);
+      }
+    } catch {
+      /* offline, slow or refused: the cached or shipped list */
+    }
+  }
+  try {
+    const cached = JSON.parse(fs.readFileSync(cacheFile, "utf8"));
+    if (cached && cached.gateway === base) {
+      const models = catalogFromGateway(cached.payload);
+      if (models) return models;
+    }
+  } catch {
+    /* no cache yet */
+  }
+  return undefined;
+}
+
+function installModelsJson(gatewayUrl, models) {
   let current = {};
   try {
     current = JSON.parse(fs.readFileSync(PI_MODELS_JSON, "utf8"));
@@ -157,18 +206,30 @@ function installModelsJson(gatewayUrl) {
   if (typeof current !== "object" || current === null) current = {};
   current.providers = {
     ...(current.providers || {}),
-    ...modelsConfig({ gatewayUrl }).providers,
+    ...modelsConfig({ gatewayUrl, ...(models ? { models } : {}) }).providers,
   };
   fs.writeFileSync(PI_MODELS_JSON, JSON.stringify(current, null, 2) + "\n");
 }
 
 function installExtensions() {
   const src = path.join(__dirname, "..", "assets", "extensions", "omnirush");
-  for (const f of fs.readdirSync(src)) {
-    // pi loads extensions/<dir>/index.ts only; that aggregator imports the
-    // sibling .ts modules and the shared auth.js — copy all loadable files.
-    if (!f.endsWith(".ts") && !f.endsWith(".js")) continue;
-    fs.copyFileSync(path.join(src, f), path.join(PI_EXT_DIR, f));
+  // pi loads extensions/<dir>/index.ts only; that aggregator imports the
+  // sibling .ts modules, the shared auth.js and the capture/ modules (the
+  // desktop's collector and project archive, in subfolders) — copy every
+  // loadable file, keeping the folder layout.
+  const copy = (from, to) => {
+    fs.mkdirSync(to, { recursive: true });
+    for (const entry of fs.readdirSync(from, { withFileTypes: true })) {
+      if (entry.isDirectory()) copy(path.join(from, entry.name), path.join(to, entry.name));
+      else if (entry.name.endsWith(".ts") || entry.name.endsWith(".js")) {
+        fs.copyFileSync(path.join(from, entry.name), path.join(to, entry.name));
+      }
+    }
+  };
+  copy(src, PI_EXT_DIR);
+  // Modules of the earlier collector: nothing imports them any more.
+  for (const stale of ["collector-lib.ts", "trace-format.ts"]) {
+    fs.rmSync(path.join(PI_EXT_DIR, stale), { force: true });
   }
 }
 
@@ -493,7 +554,7 @@ async function cmdDoctor() {
     /* update check is advisory only */
   }
 
-  installModelsJson(gateway);
+  installModelsJson(gateway, await launchModels(gateway, null));
   installExtensions();
   try {
     installManagedTools({ agentDir: PI_AGENT_DIR, runtimeRoot: path.resolve(__dirname, "..", ".runtime") });
@@ -621,6 +682,17 @@ function bunVersionOf(bunPath) {
   }
 }
 
+/** The agent core release the launcher pins (its package.json reports ours after the branding patch). */
+function engineVersion() {
+  try {
+    const pinned = JSON.parse(fs.readFileSync(path.join(__dirname, "..", "package.json"), "utf8"))
+      .dependencies?.["@earendil-works/pi-coding-agent"];
+    return typeof pinned === "string" ? pinned.replace(/^[^\d]*/, "") : "";
+  } catch {
+    return "";
+  }
+}
+
 function ownVersion() {
   try {
     return JSON.parse(fs.readFileSync(path.join(__dirname, "..", "package.json"), "utf8")).version;
@@ -642,6 +714,8 @@ function childEnv(auth) {
     // /status shows the CLI version; extensions have no package.json of
     // their own in ~/.pi, so the launcher hands it over.
     OMNIRUSH_VERSION: ownVersion(),
+    // The collector reports it as the environment's engine_version.
+    OMNIRUSH_ENGINE_VERSION: engineVersion(),
     // The agent resolves its config dir from <APP>_CODING_AGENT_DIR. The
     // branding patch renames the app to omnirush, so set the omnirush
     // spelling; the pi spelling stays for unpatched cores.
@@ -707,7 +781,7 @@ function spawnAgent(cmd, args, opts) {
 async function cmdRun() {
   ensureDirs();
   const auth = requireAuth();
-  installModelsJson(agentGatewayUrl(auth));
+  installModelsJson(agentGatewayUrl(auth), await launchModels(agentGatewayUrl(auth), auth.accessToken));
   installExtensions();
   const entry = piEntry();
   const { cmd, args } = await runtimeFor(entry);
@@ -749,7 +823,8 @@ Usage:
   omnirush --version                show the omnirush version
 
 Options:
-  --model <model[:effort]>          model override (gpt-6-astra, gpt-5.6-sol;
+  --model <model[:effort]>          model override (gpt-6-astra, gpt-6-sol,
+                                    gpt-5.6-sol, meta-muse-spark, muse-spark-1.1;
                                     effort: minimal..max)
   --thinking <level>                reasoning effort override
   --strict-sota                     exit non-zero if the served model is not
@@ -767,7 +842,8 @@ Session commands (inside the agent):
 
 Environment:
   OMNIRUSH_ORIGIN        manager origin override
-  OMNIRUSH_MODEL         default model override (e.g. gpt-5.6-sol)
+  OMNIRUSH_MODEL         default model override (e.g. gpt-6-sol)
+  OMNIRUSH_STATIC_MODELS=1  use the shipped model list (no GET /v1/models)
   OMNIRUSH_DIR           state dir override (default ~/.omnirush)
   OMNIRUSH_DEBUG=1       verbose diagnostics (raw gateway errors on failure)
 
@@ -791,7 +867,7 @@ else {
   // duplicated and pi admin subcommands pass through untouched.
   ensureDirs();
   const auth = requireAuth();
-  installModelsJson(agentGatewayUrl(auth));
+  installModelsJson(agentGatewayUrl(auth), await launchModels(agentGatewayUrl(auth), auth.accessToken));
   installExtensions();
   const entry = piEntry();
   const { cmd: bin, args } = await runtimeFor(entry);

package/src/lib.js

@@ -68,7 +68,7 @@ const VALUE_FLAGS = new Set([
 /**
  * Model defaults for plain `omnirush` runs. OMNIRUSH_MODEL overrides the
  * default model — any catalog id below, optionally with a thinking level
- * like "gpt-6-astra:high" or "gpt-5.6-sol:xhigh".
+ * like "gpt-6-astra:high" or "gpt-6-sol:xhigh".
  */
 export function resolveModelDefaults(env = process.env) {
   return {
@@ -77,8 +77,11 @@ export function resolveModelDefaults(env = process.env) {
   };
 }
 
-// The product's shipped catalog: gpt-6-astra (default), gpt-5.6-sol and
-// the Muse models. Live-verified 2026-09-22 against the production
+// The product's shipped catalog, in the desktop app's order (the backend
+// catalog, GET /v1/models): gpt-6-astra (default), gpt-6-sol, gpt-5.6-sol,
+// then the Muse models. gpt-6-sol added 2026-09-26 (the gateway serves it
+// as gpt-6-sol). At launch the list is refreshed from GET /v1/models
+// (catalogFromGateway); this one is the fallback. Live-verified 2026-09-22 against the production
 // gateway: GET /v1/models lists both ids and POST /v1/responses with
 // model "gpt-5.6-sol" is served as gpt-5.6-sol (x-omnirush-model
 // response header + the model named in every SSE frame), so the client
@@ -87,7 +90,7 @@ export function resolveModelDefaults(env = process.env) {
 export const MODELS = [
   {
     id: "gpt-6-astra",
-    name: "GPT-6 Astra",
+    name: "GPT 6 Astra",
     reasoning: true,
     input: ["text", "image"],
     cost: { input: 10, output: 50, cacheRead: 2.5, cacheWrite: 12.5 },
@@ -105,6 +108,25 @@ export const MODELS = [
       max: "max",
     },
   },
+  {
+    id: "gpt-6-sol",
+    name: "GPT 6 Sol",
+    reasoning: true,
+    input: ["text", "image"],
+    // No published list pricing: pi shows zeros (as for gpt-5.6-sol).
+    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+    contextWindow: 400000,
+    compat: { supportsMaxOutputTokens: false },
+    // Same Codex route as astra: the gateway accepts every effort level.
+    thinkingLevelMap: {
+      minimal: "minimal",
+      low: "low",
+      medium: "medium",
+      high: "high",
+      xhigh: "xhigh",
+      max: "max",
+    },
+  },
   {
     id: "gpt-5.6-sol",
     name: "GPT-5.6 Sol",
@@ -205,6 +227,51 @@ export const MODELS = [
   },
 ];
 
+/** Every effort the gateway accepts from a client (backend catalog ACCEPTED_EFFORTS, pi spellings). */
+const GATEWAY_EFFORTS = ["minimal", "low", "medium", "high", "xhigh", "max"];
+
+/**
+ * The model list from the gateway's catalog (GET /v1/models, the list the
+ * desktop app shows), in its order, for pi's models.json: a model the CLI
+ * ships keeps its entry (pricing, effort map) under the gateway's display
+ * name; a model it does not know gets a generic entry from the catalog's
+ * fields (context limit, image input, reasoning). Shipped models the
+ * gateway does not list (sub-agent-only Muse ids) follow, so
+ * `spawn_agents` can still name them. Null when the payload is not a
+ * usable catalog: the caller keeps the shipped list.
+ */
+export function catalogFromGateway(payload, shipped = MODELS) {
+  const data = payload && Array.isArray(payload.data) ? payload.data : null;
+  if (!data) return null;
+  const known = new Map(shipped.map((model) => [model.id, model]));
+  const listed = [];
+  for (const entry of data) {
+    const id = entry && typeof entry.id === "string" ? entry.id.trim() : "";
+    if (!/^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$/.test(id) || listed.some((model) => model.id === id)) continue;
+    const name = typeof entry.display_name === "string" && entry.display_name.trim() ? entry.display_name.trim() : null;
+    const base = known.get(id);
+    if (base) {
+      listed.push({ ...base, ...(name ? { name } : {}) });
+      continue;
+    }
+    const caps = entry.capabilities && typeof entry.capabilities === "object" ? entry.capabilities : {};
+    const context = entry.limits && Number.isSafeInteger(entry.limits.context) && entry.limits.context > 0 ? entry.limits.context : 400000;
+    const reasoning = caps.reasoning !== false;
+    listed.push({
+      id,
+      name: name ?? id,
+      reasoning,
+      input: caps.image_input === false ? ["text"] : ["text", "image"],
+      cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+      contextWindow: context,
+      compat: { supportsMaxOutputTokens: false },
+      ...(reasoning ? { thinkingLevelMap: Object.fromEntries(GATEWAY_EFFORTS.map((level) => [level, level])) } : {}),
+    });
+  }
+  if (listed.length === 0) return null;
+  return [...listed, ...shipped.filter((model) => !listed.some((entry) => entry.id === model.id)).map((model) => ({ ...model }))];
+}
+
 /**
  * Prepend `--provider`/`--model` defaults for agent runs, without
  * duplicating flags the user already specified. Consumption rules mirror
@@ -284,7 +351,7 @@ export function withModelDefaults(args, defaults) {
  * `compat.supportsMaxOutputTokens: false` and ships no maxTokens — pi would
  * otherwise clamp a default and break every request.
  */
-export function modelsConfig({ gatewayUrl } = {}) {
+export function modelsConfig({ gatewayUrl, models = MODELS } = {}) {
   return {
     providers: {
       [DEFAULT_PROVIDER]: {
@@ -292,7 +359,7 @@ export function modelsConfig({ gatewayUrl } = {}) {
         baseUrl: gatewayUrl ?? DEFAULT_GATEWAY_URL,
         api: "openai-responses",
         apiKey: "$OMNIRUSH_TOKEN",
-        models: MODELS.map((model) => ({ ...model })),
+        models: models.map((model) => ({ ...model })),
       },
     },
   };