cursor-route 0.1.11 → 0.1.13

package/CHANGELOG.md

@@ -4,6 +4,23 @@
 
 _(none)_
 
+## 0.1.13 — 2026-08-29
+
+- OpenRouter live-pick: do **not** cache the fetch-fail fallback (`openrouter/free`) — health no longer presents a stale fallback as `now openrouter/free`. Health says `cached <id>` only for a ranked catalog hit; otherwise `no catalog cache, fetch-fail fallback openrouter/free`.
+- Catalog fetch is an unauthenticated `GET /models` (no `Authorization` on curl argv).
+- Invalid `CURSOR_ROUTE_DS_MODEL` / `ANTHROPIC_MODEL` names those env vars instead of saying `Invalid --model`.
+- Vision auto-pick uses `\bimage\b` so "ImageMagick" stays Flash.
+- Hero demo fixture regenerated to **0.1.13**.
+
+## 0.1.12 — 2026-08-29
+
+- **Flash-first:** `--model flash` is the cheap mid default even when Grok **usage** is out. `--model pro` is harder mid / **hard backup only**, not the default Grok-out stand-in. Grok **auth** (`grok login` / `XAI_API_KEY`) still ≠ usage-out.
+- `--model vision` (and `deepseek-v4-flash-vision-exp`) on `claude-ds` / `--worker deepseek`. Mid default stays Flash. If `--model` and `CURSOR_ROUTE_DS_MODEL` / `ANTHROPIC_MODEL` are unset, screenshot/image/png/jpg/jpeg/webp/ui mock/multimodal/vision prompts auto-pick vision Flash. Explicit `--model flash|pro|vision` always wins.
+- Easy lane **live OpenRouter free pick** at request time (`GET /models`, rank `:free` or $0 text models, cache ~15 min). Do not hardcode a third-party model id as the default. Pin with `CURSOR_ROUTE_OPENROUTER_MODEL` or `--model provider/model`. Empty / `free` = live pick. Fetch-fail / empty catalog falls back to the OpenRouter **router** `openrouter/free` (fallback only). Health never fetches (cache or fallback only). Tests: `CURSOR_ROUTE_OR_OFFLINE=1` or `CURSOR_ROUTE_OR_CATALOG_JSON`.
+- `route-orch`: ProgRouter (step-wise re-route) + MoRe (do not auto-spawn N panes for multi-perspective). Flash-first table + live OpenRouter wording.
+- Hero demo fixture regenerated (`docs/fixtures/generate-hero-demo.sh`) so `--version` / health banner are 0.1.12.
+- Headless `kill`: if SIGKILL was sent, succeed even when sandboxed `ps` (EPERM) treats unreaped zombies as alive. `refreshStatus` still fail-safes to alive.
+
 ## 0.1.11 — 2026-08-21
 
 - `--worker opencode --model free` now **ranks the live OpenCode Zen catalog** (`GET https://opencode.ai/zen/v1/models`, cached ~15 min) instead of hardcoding `opencode/big-pickle`. Ox Alpha (`opencode/x-preview-f-free`) wins while it is listed and free; coding `-free` models next; `big-pickle` last among non-contributor free. Fetch-fail fallback is Ox Alpha.

package/CONTRIBUTING.md

@@ -10,6 +10,7 @@ bun run typecheck
 CURSOR_ROUTE_RELAXED=1 ./bin/cursor-route health   # without tmux
 ./bin/cursor-route start --dry-run --worker grok "ping"
 ./bin/cursor-route start --dry-run --lane mid --model flash "ping"
+./bin/cursor-route start --dry-run --lane mid --model vision "ping"
 ./bin/cursor-route start --dry-run --lane mid --model pro "ping"
 ./bin/cursor-route start --dry-run --worker deepseek --model flash "ping"   # needs dsh + DEEPSEEK_API_KEY for real starts
 ./bin/cursor-route start --dry-run --worker opencode --model free "ping"    # needs opencode for real starts
@@ -23,7 +24,7 @@ Keep `skills/route-orch/SKILL.md` and `.cursor/skills/route-orch/SKILL.md` ident
 - Do not vendor private `agent-toolkit` paths
 - Prefer adapters over rewriting the job core
 - Always-approve defaults stay documented + opt-out (`--ask` / `CURSOR_ROUTE_ASK=1`)
-- Mid default is Flash; document Pro only as harder mid / Grok **usage** stand-in (not missing `grok login`)
+- Mid default is Flash first; Pro = harder mid / **hard backup only**. Grok **auth** (`grok login`) ≠ usage-out. Do not document Pro as the default Grok-out stand-in.
 - OpenCode is opt-in (`--worker opencode`); do not make it a lane default — mid stays claude-ds
 
 ## Release checklist
@@ -31,7 +32,7 @@ Keep `skills/route-orch/SKILL.md` and `.cursor/skills/route-orch/SKILL.md` ident
 1. Update `package.json` version + [CHANGELOG.md](./CHANGELOG.md) + [docs/briefs/WORKING.md](./docs/briefs/WORKING.md)
 2. `bun test` && `bun run typecheck` && `bun run build`
 3. `cursor-route health` (tmux + ≥1 worker)
-4. One real `start` smoke (grok or claude-ds); dry-run both `--model flash` and `--model pro`
+4. One real `start` smoke (grok or claude-ds); dry-run `--model flash`, `--model vision`, and `--model pro`
 5. `npm pack --dry-run` — no secrets, no `node_modules`
 6. Commit, tag `v0.x.y`, push
 7. `npm publish --access public` with an isolated `--userconfig` Automation token (Bypass 2FA); **revoke** any chat-pasted token after

package/README.md

@@ -27,8 +27,8 @@ If you already live in Cursor, X Premium (Grok CLI), and DeepSeek — stop payin
 |------|----------------|------------------|
 | Cursor | Premium plan | Plan / synthesize / verify (orchestrator) |
 | Grok CLI | X Premium | `--lane hard` implement |
-| DeepSeek via claude-ds | DeepSeek API / plan | `--lane mid` implement (**Flash** default; `--model pro` when needed) |
-| OpenRouter free models | OpenRouter API (free tier) | `--lane easy` wording/drafts — non-secret prompts only (see Security) |
+| DeepSeek via claude-ds | DeepSeek API / plan | `--lane mid` implement (**Flash** default; `--model vision` for screenshots; `--model pro` harder mid / hard backup only) |
+| OpenRouter free models | OpenRouter API (free tier) | `--lane easy` live free pick at start — non-secret prompts only (see Security) |
 | OpenCode (opt-in) | OpenCode Zen free models | `--worker opencode` implement on live Zen free pick (Ox Alpha while listed) |
 | Codex / extra Claude | Optional | Not required for v0 |
 
@@ -116,10 +116,10 @@ docs/fixtures/generate-hero-demo.sh      # regenerate docs/fixtures/hero-demo.lo
 
 ```text
 $ cursor-route --version
-0.1.9
+0.1.13
 
 $ CURSOR_ROUTE_RELAXED=1 cursor-route health
-cursor-route v0.1.9
+cursor-route v0.1.13
 health: OK
 
 $ cursor-route start --lane mid --model flash --dry-run "Add a unit test for shellQuote"
@@ -149,18 +149,22 @@ Always-approve is **on** by default for **coding worktrees only**. It does not a
 
 | Flag | Model id | When |
 |------|----------|------|
-| `--model flash` (default) | `deepseek-v4-flash` | Cheap mid execute |
-| `--model pro` | `deepseek-v4-pro` | Harder mid / Grok **usage** stand-in |
+| `--model flash` (default) | `deepseek-v4-flash` | Cheap mid execute. Prefer this when Grok **usage** is out |
+| `--model vision` | `deepseek-v4-flash-vision-exp` | Screenshots / ui mocks / image prompts (or auto-pick) |
+| `--model pro` | `deepseek-v4-pro` | Harder mid / **hard backup** only — not the default Grok-out stand-in |
 | `--model deepseek-v4-pro[1m]` | `deepseek-v4-pro[1m]` | Large-context Pro (SKU preserved) |
 
 ```bash
 cursor-route start --lane mid "…"                    # Flash
-cursor-route start --lane mid --model pro "…"        # Pro
+cursor-route start --lane mid --model vision "…"     # Vision Flash
+cursor-route start --lane mid --model pro "…"        # Pro (harder mid / hard backup)
 # Or set default without a flag:
-export CURSOR_ROUTE_DS_MODEL=pro   # also honors ANTHROPIC_MODEL; --model overrides
+export CURSOR_ROUTE_DS_MODEL=flash   # also honors ANTHROPIC_MODEL; --model overrides
 ```
 
-**Grok auth ≠ usage-out:** if `cursor-route health` shows `worker:grok` ✗, run `grok login` (or set `XAI_API_KEY`). That is auth. Quota / subscription usage exhausted is different — use `--lane mid --model pro` as the stand-in, not a missing login.
+If `--model` and `CURSOR_ROUTE_DS_MODEL` / `ANTHROPIC_MODEL` are unset, a prompt that looks like a screenshot/image/png/jpg/jpeg/webp/ui mock/multimodal/vision auto-picks vision Flash. Explicit `--model flash|pro|vision` always wins.
+
+**Grok auth ≠ usage-out:** if `cursor-route health` shows `worker:grok` ✗, run `grok login` (or set `XAI_API_KEY`). That is auth, not the Pro case. When Grok **usage** is exhausted, stay on `--lane mid --model flash` (cheap default). `--model pro` is harder mid / hard backup only.
 
 `CURSOR_ROUTE_ALLOW_ANTHROPIC=1` is an expensive escape hatch: it does **not** pass DeepSeek `--model` ids (Anthropic would reject them).
 
@@ -180,8 +184,9 @@ export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-flash
 export CLAUDE_CODE_SUBAGENT_MODEL=deepseek-v4-flash
 
 cursor-route health          # worker:claude-ds should be ✓; `lane:mid` is ✓ only when DeepSeek is proven
-cursor-route start --lane mid "…"                 # Flash (default)
-cursor-route start --lane mid --model pro "…"     # Pro when you need it
+cursor-route start --lane mid "…"                 # Flash (default; also when Grok usage is out)
+cursor-route start --lane mid --model vision "…"  # screenshots / ui mocks
+cursor-route start --lane mid --model pro "…"     # harder mid / hard backup only
 ```
 
 Persist the same vars under `~/.claude/settings.json` → `"env": { … }` if you want them every shell.
@@ -196,10 +201,10 @@ export DEEPSEEK_API_KEY=YOUR_DEEPSEEK_API_KEY   # from platform.deepseek.com
 
 cursor-route health                              # worker:deepseek should be ✓
 cursor-route start --worker deepseek "…"         # Flash (default)
-cursor-route start --worker deepseek --model pro "…"   # Pro when you need it
+cursor-route start --worker deepseek --model pro "…"   # harder mid / hard backup only
 ```
 
-The adapter launches `dsh --profile headless` with a **per-job Cordis patch** (`jobs/<id>.dsh-patch.yml`, mode 0600) that pins `--model flash|pro` (`deepseek-v4-pro[1m]` preserved) — it never rewrites `~/.dsh/settings.yaml`, so parallel jobs don't race. Always-approve maps to `DSH_PERMISSION_MODE=danger-full-access`; `--ask` drops to `workspace-write`. Your `DEEPSEEK_API_KEY` travels via env only — never in the launch command or patch. Override the binary with `CURSOR_ROUTE_DSH_BIN`.
+The adapter launches `dsh --profile headless` with a **per-job Cordis patch** (`jobs/<id>.dsh-patch.yml`, mode 0600) that pins `--model flash|pro|vision` (`deepseek-v4-pro[1m]` preserved) — it never rewrites `~/.dsh/settings.yaml`, so parallel jobs don't race. Always-approve maps to `DSH_PERMISSION_MODE=danger-full-access`; `--ask` drops to `workspace-write`. Your `DEEPSEEK_API_KEY` travels via env only — never in the launch command or patch. Override the binary with `CURSOR_ROUTE_DSH_BIN`.
 
 **Not the default:** bare `claude` still talking to Anthropic. Health refuses that so a misconfigured install cannot silently burn frontier $ rates. Escape hatch only: `CURSOR_ROUTE_ALLOW_ANTHROPIC=1`.
 
@@ -207,19 +212,27 @@ No DeepSeek yet? Use `--lane hard` / `--worker grok` (X Premium).
 
 ## OpenRouter setup (the free easy lane)
 
-`--lane easy` / `--worker openrouter` sends wording/draft prompts to OpenRouter's
-free model route (`openrouter/free`). Get a key at [openrouter.ai/keys](https://openrouter.ai/keys).
+`--lane easy` / `--worker openrouter` sends wording/draft prompts to OpenRouter
+and **live-picks the best free text model** at request time (`GET /models`, rank
+`:free` or $0 text models). Do not hardcode a specific model id as the default.
+Pin with `CURSOR_ROUTE_OPENROUTER_MODEL` or `--model provider/model`. Empty /
+`free` = live pick. If the catalog fetch fails, the fallback is OpenRouter's
+**router** id `openrouter/free` (a live router, not a locked model). Get a key
+at [openrouter.ai/keys](https://openrouter.ai/keys).
 
 ```bash
 export OPENROUTER_API_KEY=sk-or-v1-...        # from openrouter.ai/keys
-# optional:
-export CURSOR_ROUTE_OPENROUTER_MODEL=openrouter/free   # default
+# optional pin (skips the live catalog pick):
+export CURSOR_ROUTE_OPENROUTER_MODEL=qwen/qwen3-coder:free
 export OPENROUTER_BASE_URL=https://openrouter.ai/api/v1  # default
 
-cursor-route health          # worker:openrouter should be ✓
+cursor-route health          # worker:openrouter should be ✓ (never fetches /models)
 cursor-route start --lane easy "Rewrite this FAQ answer in 3 sentences"
+cursor-route start --lane easy --model free "…"          # live pick
 ```
 
+`health` never fetches the OpenRouter catalog (cache or fallback only), same as Zen.
+
 **Non-secret prompts only:** free OpenRouter models may log prompts, so the easy lane is for
 **wording/drafts without credentials**. The same refuse gate as every lane blocks
 key-shaped material in `start` / `send`, and the runner re-checks the prompt file.
@@ -277,13 +290,13 @@ cursor-route is a public MIT CLI and Cursor skill that runs parallel coding work
 It uses the familiar strategist and worker-pane shape, but it is not a Codex clone. cursor-route uses Cursor as the planner and DeepSeek plus Grok CLI as workers. Codex is not required.
 
 **Does mid lane use Anthropic Claude?**  
-No. The mid worker is DeepSeek. Claude Code is the harness, configured with `ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic` and a DeepSeek key in `ANTHROPIC_AUTH_TOKEN`. Default model is Flash (`--model flash`); use `--model pro` for harder mid work or when Grok usage is exhausted (not the same as a missing `grok login`).
+No. The mid worker is DeepSeek. Claude Code is the harness, configured with `ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic` and a DeepSeek key in `ANTHROPIC_AUTH_TOKEN`. Default model is Flash (`--model flash`) — keep Flash when Grok **usage** is out. `--model vision` is for screenshots / ui mocks. `--model pro` is harder mid / **hard backup only**, not the default Grok-out stand-in. A missing `grok login` is auth, not the Pro case.
 
 **How do I install?**  
 Run `npm i -g cursor-route`, install tmux if needed, then run `cursor-route health`. The package is available at https://www.npmjs.com/package/cursor-route, and the source is at https://github.com/cemini23/cursor-route.
 
 **Is it free?**  
-The cursor-route code is open source under MIT. It does not make the worker services free. Your costs depend on Cursor, DeepSeek API usage, and the Grok access or balance available to you. The easy lane can be free on OpenRouter's free-model route (`openrouter/free`). `--worker opencode` can run live OpenCode Zen free models (`--model free` ranks the catalog).
+The cursor-route code is open source under MIT. It does not make the worker services free. Your costs depend on Cursor, DeepSeek API usage, and the Grok access or balance available to you. The easy lane live-picks a free OpenRouter text model at start (fallback router `openrouter/free` only if the catalog fetch fails). `--worker opencode` can run live OpenCode Zen free models (`--model free` ranks the catalog).
 
 ## Related
 

package/dist/adapters/claude-ds.js

@@ -191,7 +191,7 @@ export const claudeDsAdapter = {
         const isDeepSeekStock = resolved.mode.startsWith("claude → DeepSeek");
         const isShim = resolved.mode.startsWith("claude-ds") || resolved.mode.startsWith("deepseek-claude");
         // Anthropic escape hatch: do not pass DeepSeek model ids (unknown to Anthropic).
-        // --model flash|pro is DeepSeek-only; stock Claude uses its own defaults / ANTHROPIC_MODEL.
+        // --model flash|pro|vision is DeepSeek-only; stock Claude uses its own defaults / ANTHROPIC_MODEL.
         if (isAnthropicEscape) {
             const parts = [
                 shellQuote(resolved.binary),

package/dist/adapters/openrouter.js

@@ -2,7 +2,7 @@ import { existsSync } from "node:fs";
 import { fileURLToPath } from "node:url";
 import { dirname, join } from "node:path";
 import { shellQuote } from "../util.js";
-import { openRouterModel, openRouterBaseUrl } from "../config.js";
+import { cachedOrFreePick, OPENROUTER_FALLBACK_MODEL, openRouterBaseUrl, } from "../config.js";
 /**
  * Resolve how to invoke the one-shot runner. Prefer the compiled dist via node
  * (no loader); else Bun on src. No npx/tsx — same policy as mark-complete.
@@ -19,12 +19,12 @@ function resolveRunner() {
     }
     return null;
 }
-function openRouterEnv() {
+function openRouterEnv(modelId) {
     const key = process.env.OPENROUTER_API_KEY;
     if (!key)
         return undefined;
     const env = { OPENROUTER_API_KEY: key };
-    const model = process.env.CURSOR_ROUTE_OPENROUTER_MODEL;
+    const model = modelId || process.env.CURSOR_ROUTE_OPENROUTER_MODEL;
     if (model)
         env.CURSOR_ROUTE_OPENROUTER_MODEL = model;
     const base = process.env.OPENROUTER_BASE_URL;
@@ -42,7 +42,7 @@ export const openRouterAdapter = {
                 worker: "openrouter",
                 ok: false,
                 binary: runner?.command ?? null,
-                detail: "OPENROUTER_API_KEY not set — export your OpenRouter key (easy lane model defaults to openrouter/free)",
+                detail: "OPENROUTER_API_KEY not set — export your OpenRouter key (easy lane live-picks a free model at start; pin with CURSOR_ROUTE_OPENROUTER_MODEL)",
             };
         }
         if (!runner) {
@@ -53,20 +53,26 @@ export const openRouterAdapter = {
                 detail: "openrouter-run not found — run bun run build (or use Bun from a source clone)",
             };
         }
+        // Health stays offline: ranked cache hit, else label the router fallback
+        // (do not present openrouter/free as a fresh live pick).
+        const cached = cachedOrFreePick();
+        const detail = cached
+            ? `ok (live pick at start; cached ${cached} @ ${openRouterBaseUrl()})`
+            : `ok (live pick at start; no catalog cache, fetch-fail fallback ${OPENROUTER_FALLBACK_MODEL} @ ${openRouterBaseUrl()})`;
         return {
             worker: "openrouter",
             ok: true,
             binary: runner.command,
-            detail: `ok (model ${openRouterModel()} @ ${openRouterBaseUrl()})`,
+            detail,
         };
     },
-    buildLaunch({ promptFile }) {
+    buildLaunch({ promptFile, modelId }) {
         const runner = resolveRunner();
         if (!runner)
             throw new Error("openrouter runner not available — run: bun run build");
         // Missing key is tolerated here so `--dry-run` can still print the command;
         // real starts are gated by the health preflight (which requires the key).
-        const env = openRouterEnv();
+        const env = openRouterEnv(modelId);
         // No interactive approval concept for a pure HTTP call — nothing to auto-approve.
         return {
             worker: "openrouter",

package/dist/cli.js

@@ -4,7 +4,7 @@
  */
 import { readFileSync, existsSync, realpathSync, statSync } from "node:fs";
 import { resolve, basename } from "node:path";
-import { config, WORKERS, LANES, resolveDsModel, openCodeModel, } from "./config.js";
+import { config, WORKERS, LANES, resolveDsModel, openCodeModel, openRouterModel, } from "./config.js";
 import { runHealth, printHealth } from "./health.js";
 import { startJob, listJobs, readJob, killJob, cleanJobs, jobPaths, refreshStatus, jobEvidence, } from "./jobs.js";
 import { capturePane, sendKeys, attachHint, listManagedSessions, sessionExists, } from "./tmux.js";
@@ -31,7 +31,7 @@ Usage:
 Start options:
   --worker <grok|claude-ds|openrouter|deepseek|opencode>  Worker adapter (default: grok; deepseek/opencode = opt-in)
   --lane <easy|mid|hard>                Lane → worker (easy=openrouter, mid=claude-ds, hard=grok)
-  --model <flash|pro|free|provider/model>  claude-ds/deepseek: flash|pro. opencode: free (live Zen catalog pick) or provider/model
+  --model <flash|pro|vision|free|provider/model>  claude-ds/deepseek: flash|pro|vision. opencode/openrouter: free (live catalog pick) or provider/model
   --dir <path>                          Working directory (default: cwd)
   --ask                                 Disable always-approve for this job
   --dry-run                             Print launch command; do not start
@@ -44,14 +44,14 @@ Env:
   CURSOR_ROUTE_MAX_JOBS              Max active jobs (default: 50)
   CURSOR_ROUTE_RELAXED=1             health OK without tmux/workers (CI / infra smoke)
   CURSOR_ROUTE_ALLOW_ANTHROPIC=1     Allow mid-lane on Anthropic Claude (expensive; not default; --model ignored)
-  CURSOR_ROUTE_DS_MODEL              Default mid model flash|pro (or deepseek-v4-pro[1m]); overridden by --model
+  CURSOR_ROUTE_DS_MODEL              Default mid model flash|pro|vision (or full ids / deepseek-v4-pro[1m]); overridden by --model
   CURSOR_ROUTE_GROK_BIN              Override the grok binary path (tests / power users)
   CURSOR_ROUTE_CLAUDE_DS_BIN         Override the claude-ds binary path (tests / power users)
   CURSOR_ROUTE_DSH_BIN               Override the dsh binary path (tests / power users)
   CURSOR_ROUTE_OPENCODE_BIN          Override the opencode binary path (tests / power users)
   DEEPSEEK_API_KEY                   DeepSeek API key (required for --worker deepseek)
   OPENROUTER_API_KEY                 OpenRouter key (required for --worker openrouter / --lane easy)
-  CURSOR_ROUTE_OPENROUTER_MODEL      OpenRouter model (default: openrouter/free)
+  CURSOR_ROUTE_OPENROUTER_MODEL      OpenRouter pin (unset or free = live catalog pick at start; fetch-fail fallback openrouter/free)
   OPENROUTER_BASE_URL                OpenRouter API base (default: https://openrouter.ai/api/v1)
   CURSOR_ROUTE_OPENCODE_MODEL        OpenCode model pin (unset or free = live Zen free pick); --model overrides
 `);
@@ -149,7 +149,7 @@ function asDsModelChoice(v) {
     if (v === undefined || v === true)
         return undefined;
     if (typeof v !== "string")
-        throw new Error(`Invalid --model; expected flash|pro`);
+        throw new Error(`Invalid --model; expected flash|pro|vision`);
     return resolveDsModel(v);
 }
 function asOpenCodeModel(v) {
@@ -160,6 +160,14 @@ function asOpenCodeModel(v) {
     }
     return openCodeModel(v);
 }
+function asOpenRouterModel(v) {
+    if (v === undefined || v === true)
+        return undefined;
+    if (typeof v !== "string") {
+        throw new Error(`Invalid --model; expected provider/model or free (live OpenRouter catalog pick)`);
+    }
+    return openRouterModel(v);
+}
 function refuseSecrets(text, context) {
     if (looksLikeSecretMaterial(text)) {
         console.error(`Refusing ${context}: looks like secret key material. Remove tokens/keys and retry.`);
@@ -233,8 +241,8 @@ async function main() {
         const resolvedWorker = worker ?? (lane ? config.laneWorkers[lane] : config.defaultWorker);
         let model;
         let modelId;
-        // --model: DeepSeek flash|pro for claude-ds/deepseek; provider/model (or free) for opencode;
-        // ignore (do not validate) for grok/openrouter
+        // --model: DeepSeek flash|pro|vision for claude-ds/deepseek;
+        // provider/model (or free) for opencode/openrouter; ignore for grok
         if (f.model !== undefined && (resolvedWorker === "claude-ds" || resolvedWorker === "deepseek")) {
             try {
                 const choice = asDsModelChoice(f.model);
@@ -257,6 +265,15 @@ async function main() {
                 process.exit(2);
             }
         }
+        else if (f.model !== undefined && resolvedWorker === "openrouter") {
+            try {
+                modelId = asOpenRouterModel(f.model);
+            }
+            catch (e) {
+                console.error(e.message);
+                process.exit(2);
+            }
+        }
         let prompt = "";
         if (promptFile) {
             const p = resolve(promptFile);

package/dist/config.js

@@ -3,11 +3,17 @@ import { join } from "node:path";
 import { defaultJobsDir } from "./runtime.js";
 export const WORKERS = ["grok", "claude-ds", "openrouter", "deepseek", "opencode"];
 export const LANES = ["easy", "mid", "hard"];
-export const DS_MODELS = ["flash", "pro"];
+export const DS_MODELS = ["flash", "pro", "vision"];
 export const DS_MODEL_IDS = {
     flash: "deepseek-v4-flash",
     pro: "deepseek-v4-pro",
+    vision: "deepseek-v4-flash-vision-exp",
 };
+/** Prompt looks like a screenshot / image / ui mock — vision auto-pick. */
+const VISION_PROMPT_RE = /screenshot|\bimage\b|\.png|\.jpe?g|\.webp|ui mock|multimodal|\bvision\b/i;
+export function promptLooksLikeVision(prompt) {
+    return VISION_PROMPT_RE.test(prompt);
+}
 /**
  * Resolve --model / env to alias + concrete model id.
  * Preserves `deepseek-v4-pro[1m]` (does not silently strip the SKU).
@@ -24,10 +30,13 @@ export function resolveDsModel(raw) {
     if (v === "pro" || v === "deepseek-v4-pro") {
         return { alias: "pro", id: DS_MODEL_IDS.pro };
     }
+    if (v === "vision" || v === "deepseek-v4-flash-vision-exp") {
+        return { alias: "vision", id: DS_MODEL_IDS.vision };
+    }
     if (v === "deepseek-v4-pro[1m]") {
         return { alias: "pro", id: "deepseek-v4-pro[1m]" };
     }
-    throw new Error(`Invalid --model ${raw}; expected flash|pro`);
+    throw new Error(`Invalid DeepSeek model ${raw}; expected flash|pro|vision`);
 }
 /** Alias-only helper (tests / callers that do not need the concrete id). */
 export function resolveDsModelAlias(raw) {
@@ -41,15 +50,12 @@ export function defaultDsModelFromEnv() {
     const raw = process.env.CURSOR_ROUTE_DS_MODEL || process.env.ANTHROPIC_MODEL;
     return resolveDsModel(raw);
 }
-/** OpenRouter model for the easy lane (env CURSOR_ROUTE_OPENROUTER_MODEL). */
-export function openRouterModel() {
-    return process.env.CURSOR_ROUTE_OPENROUTER_MODEL || "openrouter/free";
-}
 /** OpenRouter API base URL (env OPENROUTER_BASE_URL). */
 export function openRouterBaseUrl() {
     return process.env.OPENROUTER_BASE_URL || "https://openrouter.ai/api/v1";
 }
 export { OPENCODE_DEFAULT_MODEL, assertOpenCodeModel, openCodeModel, cachedZenFreePick, } from "./zen-free.js";
+export { OPENROUTER_FALLBACK_MODEL, assertOpenRouterModel, cachedOrFreePick, openRouterModel, } from "./or-free.js";
 function maxConcurrentJobsFromEnv() {
     const raw = process.env.CURSOR_ROUTE_MAX_JOBS;
     if (raw) {
@@ -66,7 +72,7 @@ function maxConcurrentJobsFromEnv() {
  */
 export const config = {
     product: "cursor-route",
-    version: "0.1.11",
+    version: "0.1.13",
     get jobsDir() {
         return defaultJobsDir();
     },

package/dist/jobs.js

@@ -2,7 +2,7 @@ import { mkdirSync, writeFileSync, readFileSync, readdirSync, existsSync, unlink
 import { join, resolve } from "node:path";
 import { fileURLToPath } from "node:url";
 import { dirname } from "node:path";
-import { config, sessionName, defaultDsModelFromEnv, DS_MODEL_IDS, openCodeModel, } from "./config.js";
+import { config, sessionName, defaultDsModelFromEnv, DS_MODEL_IDS, openCodeModel, openRouterModel, promptLooksLikeVision, } from "./config.js";
 import { getAdapter } from "./adapters/index.js";
 import { spawn } from "node:child_process";
 import { spawnSync } from "node:child_process";
@@ -126,14 +126,18 @@ function terminatePid(pid) {
             return true;
         sleepMs(100);
     }
-    trySignal("SIGKILL", true);
-    trySignal("SIGKILL", false);
+    const gKill = trySignal("SIGKILL", true);
+    const pKill = trySignal("SIGKILL", false);
     for (let i = 0; i < 10; i++) {
         if (!pidAlive(pid))
             return true;
         sleepMs(100);
     }
-    return !pidAlive(pid);
+    if (!pidAlive(pid))
+        return true;
+    // Sandboxed `ps` (EPERM) treats unreaped zombies as alive. If SIGKILL was
+    // sent, killJob should succeed — refreshStatus still fail-safes to alive.
+    return gKill || pKill;
 }
 /** Refresh running jobs without inventing success from "session gone". */
 export function refreshStatus(job) {
@@ -253,20 +257,36 @@ export function startJob(opts) {
             modelId = opts.modelId ?? DS_MODEL_IDS[opts.model];
         }
         else {
-            try {
-                const choice = defaultDsModelFromEnv();
-                dsAlias = choice.alias;
-                model = choice.alias;
-                modelId = opts.modelId ?? choice.id;
-            }
-            catch (e) {
+            const envRaw = (process.env.CURSOR_ROUTE_DS_MODEL || process.env.ANTHROPIC_MODEL || "").trim();
+            if (envRaw) {
                 try {
-                    unlinkSync(paths.prompt);
+                    const choice = defaultDsModelFromEnv();
+                    dsAlias = choice.alias;
+                    model = choice.alias;
+                    modelId = opts.modelId ?? choice.id;
                 }
                 catch {
-                    /* ignore */
+                    try {
+                        unlinkSync(paths.prompt);
+                    }
+                    catch {
+                        /* ignore */
+                    }
+                    return {
+                        ok: false,
+                        error: `Invalid CURSOR_ROUTE_DS_MODEL/ANTHROPIC_MODEL value ${envRaw}; expected flash|pro|vision`,
+                    };
                 }
-                return { ok: false, error: e.message };
+            }
+            else if (promptLooksLikeVision(opts.prompt)) {
+                dsAlias = "vision";
+                model = "vision";
+                modelId = opts.modelId ?? DS_MODEL_IDS.vision;
+            }
+            else {
+                dsAlias = "flash";
+                model = "flash";
+                modelId = opts.modelId ?? DS_MODEL_IDS.flash;
             }
         }
     }
@@ -286,6 +306,22 @@ export function startJob(opts) {
             return { ok: false, error: e.message };
         }
     }
+    else if (worker === "openrouter") {
+        try {
+            const orModel = openRouterModel(opts.modelId);
+            model = orModel;
+            modelId = orModel;
+        }
+        catch (e) {
+            try {
+                unlinkSync(paths.prompt);
+            }
+            catch {
+                /* ignore */
+            }
+            return { ok: false, error: e.message };
+        }
+    }
     let plan;
     try {
         plan = adapter.buildLaunch({

package/dist/openrouter-run.js

@@ -19,7 +19,7 @@ function parseArgs(argv) {
         const a = argv[i];
         if (a === "-h" || a === "--help") {
             console.log("usage: cursor-route/openrouter-run --prompt-file <path>\n" +
-                "env: OPENROUTER_API_KEY (required), CURSOR_ROUTE_OPENROUTER_MODEL, OPENROUTER_BASE_URL");
+                "env: OPENROUTER_API_KEY (required), CURSOR_ROUTE_OPENROUTER_MODEL (unset/free = live pick), OPENROUTER_BASE_URL");
             process.exit(0);
         }
         if (a === "--prompt-file") {