ci: reference GITEA secret by its actual name

- provider-agnostic llm-client (grok + openrouter) with page-slug context
- live golden run scored and passing via x-ai/grok-4.6
- ADR 0010; spike docs flipped to done

.env.example

@@ -8,9 +8,12 @@ REDIS_URL=redis://localhost:6379
 # Auth
 SESSION_SECRET=replace-me
 
-# LLM — Grok is the default adapter. Switch provider with LLM_PROVIDER.
+# LLM — Grok is the default adapter. Switch provider with LLM_PROVIDER (grok | openrouter).
 LLM_PROVIDER=grok
 XAI_API_KEY=
+OPENROUTER_API_KEY=
+# Model sent when LLM_PROVIDER=openrouter (default x-ai/grok-4.6).
+OPENROUTER_MODEL=x-ai/grok-4.6
 LLM_FALLBACK_PROVIDER=
 
 # Sandbox

.gitea/workflows/mirror.yml

@@ -22,4 +22,4 @@ jobs:
       - name: Reconcile with GitHub
         run: bash scripts/mirror.sh
         env:
-          MIRROR_URL: https://x-access-token:${{ secrets.GITHUB_TOKEN }}@github.com/SinachPat/wursor.git
+          MIRROR_URL: https://x-access-token:${{ secrets.GITHUB_TOKEN }}@github.com/Wursor/wursor.git

.github/workflows/ci.yml

@@ -0,0 +1,27 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: pnpm/action-setup@v4
+        with:
+          version: 10.33.0
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: pnpm
+
+      - run: pnpm install --frozen-lockfile
+
+      - run: pnpm test
+
+      - run: pnpm lint

.github/workflows/mirror.yml

@@ -22,4 +22,4 @@ jobs:
       - name: Reconcile with Gitea
         run: bash scripts/mirror.sh
         env:
-          MIRROR_URL: https://x-access-token:${{ secrets.GITEA_TOKEN }}@git.weown.tools/pat/wursor.git
+          MIRROR_URL: https://x-access-token:${{ secrets.GITEA }}@git.weown.tools/pat/wursor.git

PRD.md

@@ -12,7 +12,7 @@ The Agentic WordPress Management Platform
 | **Date** | August 13, 2026 |
 | **Author** | Patrick (Product Lead) |
 | **Status** | Draft — Internal (non-technical-first pivot; Phase 0) |
-| **Repo** | SinachPat/wursor |
+| **Repo** | Wursor/wursor |
 | **Classification** | Confidential |
 | **Supersedes** | v1.3 (engineer-first; desktop shell; Tauri + Monaco) |
 

docs/decisions/0009-repo-rename.md

@@ -24,3 +24,5 @@ Rename the repository to `wursor`, set the description to the product tagline, c
 
 - Remote is `https://github.com/SinachPat/wursor`; identity matches the product.
 - Redirects from the old name are handled by GitHub automatically.
+
+> **Follow-up (2026-08-15):** the repo was later transferred to the `Wursor` GitHub organization — canonical path is now `https://github.com/Wursor/wursor`.

docs/decisions/0010-openrouter-live-golden.md

@@ -0,0 +1,30 @@
+# 10. Golden harness scores live runs through a provider-agnostic LLM client (OpenRouter first)
+
+- **Status:** Accepted
+- **Date:** 2026-08-15
+
+## Context
+
+Phase 0's golden-task spike was the last gate item, stuck at "partial" because no `XAI_API_KEY` was set and the harness hard-coded `api.x.ai`. The developer holds an OpenRouter key, and OpenRouter speaks the same OpenAI-compatible `chat/completions` shape (including tool-calling), so the live run could be unblocked without a Grok-specific key.
+
+## Decision
+
+Replace the hard-coded `grok-client.ts` with a provider-agnostic `llm-client.ts` that supports `grok` (x.ai) and `openrouter`, selected by `LLM_PROVIDER`. The OpenRouter default model is `x-ai/grok-4.6`. The harness now passes the site's page slugs into the prompt.
+
+### Options considered
+
+- Wait for an `XAI_API_KEY` and keep the hard-coded x.ai client.
+- Hard-code OpenRouter, dropping the x.ai path.
+- Provider-agnostic client with `grok` + `openrouter` (chosen).
+
+### Rejected
+
+- Wait for x.ai — blocks the gate on a key we don't have, for no technical reason.
+- OpenRouter-only — the plan (IMPLEMENTATION §3) still names Grok the default adapter; keeping both providers matches that plan and costs one env switch.
+
+## Consequences
+
+- The live gate run is scored and passing (`gb-01` passed on `x-ai/grok-4.6`); the Phase 0 golden spike flips to done.
+- `x-ai/grok-latest` is not a callable OpenRouter ID (returns 400); the pinned `x-ai/grok-4.6` works and supports `tool_choice: required`.
+- A real finding: without the page-slug list in the prompt, the model guessed `page: "home"` and failed. Tool-call prompts must always carry site/page context. Sprint 3 must inherit this when it builds `api/src/agents/llm-client.ts`.
+- `llm-client.ts` here is a spike; Sprint 3 will generalize it (streaming, circuit breaker, fallback) rather than promote this file verbatim.

docs/decisions/README.md

@@ -23,6 +23,7 @@ Each ADR is a single file following the [Nygard format](https://cognitect.com/bl
 | [0007](0007-media-proxy-not-copy.md) | Sandboxes proxy uploads; never copy the media library | Accepted |
 | [0008](0008-empty-packages-not-stubs.md) | Workspace ships empty packages, not placeholder source | Accepted |
 | [0009](0009-repo-rename.md) | Repository renamed originmain → wursor | Accepted |
+| [0010](0010-openrouter-live-golden.md) | Golden harness scores live runs through a provider-agnostic LLM client (OpenRouter first) | Accepted |
 
 ## How to add one
 

e2e/golden/__tests__/llm-client.test.ts

@@ -0,0 +1,98 @@
+import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
+import { resolveProvider, callLlm } from '../src/llm-client.ts';
+
+let fetchMock: ReturnType<typeof vi.fn>;
+
+beforeEach(() => {
+  fetchMock = vi.fn();
+  vi.stubGlobal('fetch', fetchMock);
+});
+
+afterEach(() => {
+  vi.unstubAllGlobals();
+});
+
+describe('resolveProvider', () => {
+  it('maps grok to the x.ai endpoint and grok-3 model', () => {
+    expect(resolveProvider('grok')).toEqual({ baseUrl: 'https://api.x.ai/v1', model: 'grok-3' });
+  });
+
+  it('maps openrouter to the openrouter endpoint and the grok-4.6 model', () => {
+    expect(resolveProvider('openrouter')).toEqual({
+      baseUrl: 'https://openrouter.ai/api/v1',
+      model: 'x-ai/grok-4.6',
+    });
+  });
+
+  it('honors an explicit model override', () => {
+    expect(resolveProvider('openrouter', 'openai/gpt-4o')).toEqual({
+      baseUrl: 'https://openrouter.ai/api/v1',
+      model: 'openai/gpt-4o',
+    });
+  });
+});
+
+describe('callLlm', () => {
+  it('posts an OpenAI-shaped tool-call request to the grok endpoint', async () => {
+    fetchMock.mockResolvedValue({ ok: true, json: async () => ({ choices: [] }) });
+
+    await callLlm({
+      provider: 'grok',
+      apiKey: 'test-key',
+      prompt: 'Change the heading',
+      siteId: 'gutenberg-business',
+      builder: 'gutenberg',
+    });
+
+    const [url, init] = fetchMock.mock.calls[0] as [string, { headers: Record<string, string>; body: string }];
+    expect(url).toBe('https://api.x.ai/v1/chat/completions');
+    expect(init.headers.Authorization).toBe('Bearer test-key');
+    const body = JSON.parse(init.body) as { model: string; tool_choice: string; tools: Array<{ function: { name: string } }> };
+    expect(body.model).toBe('grok-3');
+    expect(body.tool_choice).toBe('required');
+    expect(body.tools[0]?.function.name).toBe('edit_heading');
+  });
+
+  it('adds OpenRouter identification headers and uses the openrouter endpoint', async () => {
+    fetchMock.mockResolvedValue({ ok: true, json: async () => ({ choices: [] }) });
+
+    await callLlm({
+      provider: 'openrouter',
+      apiKey: 'test-key',
+      prompt: 'Change the heading',
+      siteId: 'gutenberg-business',
+      builder: 'gutenberg',
+    });
+
+    const [url, init] = fetchMock.mock.calls[0] as [string, { headers: Record<string, string>; body: string }];
+    expect(url).toBe('https://openrouter.ai/api/v1/chat/completions');
+    expect(init.headers['HTTP-Referer']).toBeDefined();
+    expect(init.headers['X-Title']).toBeDefined();
+    expect(JSON.parse(init.body).model).toBe('x-ai/grok-4.6');
+  });
+
+  it('includes page slugs in the user message when provided', async () => {
+    fetchMock.mockResolvedValue({ ok: true, json: async () => ({ choices: [] }) });
+
+    await callLlm({
+      provider: 'grok',
+      apiKey: 'test-key',
+      prompt: 'Change the heading',
+      siteId: 'gutenberg-business',
+      builder: 'gutenberg',
+      pages: ['homepage', 'about', 'contact'],
+    });
+
+    const [, init] = fetchMock.mock.calls[0] as [string, { body: string }];
+    const body = JSON.parse(init.body) as { messages: Array<{ content: string }> };
+    expect(body.messages[1]?.content).toContain('pages=homepage,about,contact');
+  });
+
+  it('throws when the provider returns a non-ok response', async () => {
+    fetchMock.mockResolvedValue({ ok: false, status: 401 });
+
+    await expect(
+      callLlm({ provider: 'grok', apiKey: 'bad', prompt: 'x', siteId: 's', builder: 'b' }),
+    ).rejects.toThrow('LLM HTTP 401');
+  });
+});

e2e/golden/runs/latest.json

@@ -2,8 +2,8 @@
   "fixturePassed": 20,
   "fixtureTotal": 20,
   "grokLive": {
-    "skipped": true,
+    "id": "gb-01",
-    "reason": "XAI_API_KEY not set"
+    "passed": true
   },
   "scores": [
     {

e2e/golden/src/llm-client.ts

@@ -1,21 +1,47 @@
 import type { GrokResponse } from './types.ts';
 
-const url = 'https://api.x.ai/v1/chat/completions';
+export type LlmProvider = 'grok' | 'openrouter';
 
-export async function callGrok(input: {
+export type LlmConfig = {
+  baseUrl: string;
+  model: string;
+};
+
+const PROVIDERS: Record<LlmProvider, LlmConfig> = {
+  grok: { baseUrl: 'https://api.x.ai/v1', model: 'grok-3' },
+  openrouter: { baseUrl: 'https://openrouter.ai/api/v1', model: 'x-ai/grok-4.6' },
+};
+
+export function resolveProvider(provider: LlmProvider, model?: string): LlmConfig {
+  const config = PROVIDERS[provider];
+  return { baseUrl: config.baseUrl, model: model ?? config.model };
+}
+
+export async function callLlm(input: {
+  provider: LlmProvider;
   apiKey: string;
+  model?: string;
   prompt: string;
   siteId: string;
   builder: string;
+  pages?: string[];
 }): Promise<GrokResponse> {
-  const response = await fetch(url, {
+  const { baseUrl, model } = resolveProvider(input.provider, input.model);
-    method: 'POST',
+
-    headers: {
+  const headers: Record<string, string> = {
     Authorization: `Bearer ${input.apiKey}`,
     'Content-Type': 'application/json',
-    },
+  };
+  if (input.provider === 'openrouter') {
+    headers['HTTP-Referer'] = 'https://wursor.dev';
+    headers['X-Title'] = 'Wursor golden harness';
+  }
+
+  const response = await fetch(`${baseUrl}/chat/completions`, {
+    method: 'POST',
+    headers,
     body: JSON.stringify({
-      model: 'grok-3',
+      model,
       messages: [
         {
           role: 'system',
@@ -24,7 +50,7 @@ export async function callGrok(input: {
         },
         {
           role: 'user',
-          content: `site=${input.siteId} builder=${input.builder}\n${input.prompt}`,
+          content: `site=${input.siteId} builder=${input.builder}${input.pages ? ` pages=${input.pages.join(',')}` : ''}\n${input.prompt}`,
         },
       ],
       tools: [
@@ -71,7 +97,7 @@ export async function callGrok(input: {
   });
 
   if (!response.ok) {
-    throw new Error(`Grok HTTP ${response.status}`);
+    throw new Error(`LLM HTTP ${response.status}`);
   }
   return (await response.json()) as GrokResponse;
 }

e2e/golden/src/run-golden.ts

@@ -1,17 +1,29 @@
 import { mkdirSync, writeFileSync } from 'node:fs';
 import { dirname, join } from 'node:path';
+import { loadEnvFile } from 'node:process';
 import { fileURLToPath } from 'node:url';
 import { detectBuilder } from './builder-detect.ts';
 import { asGrokResponse, expectedCalls } from './expected-calls.ts';
-import { callGrok } from './grok-client.ts';
+import { callLlm, type LlmProvider } from './llm-client.ts';
 import { loadPrompts } from './load-prompts.ts';
 import { loadSite } from './load-site.ts';
 import { scoreGrokResponse } from './score.ts';
 
 const goldenRoot = join(dirname(fileURLToPath(import.meta.url)), '..');
+const repoRoot = join(goldenRoot, '..', '..');
+
+try {
+  loadEnvFile(join(repoRoot, '.env'));
+} catch {
+  // no .env at repo root — live run will skip
+}
+
+function provider(): LlmProvider {
+  return process.env.LLM_PROVIDER === 'openrouter' ? 'openrouter' : 'grok';
+}
 
 function key(): string | undefined {
-  const value = process.env.XAI_API_KEY;
+  const value = provider() === 'openrouter' ? process.env.OPENROUTER_API_KEY : process.env.XAI_API_KEY;
   return value !== undefined && value !== '' ? value : undefined;
 }
 
@@ -39,11 +51,14 @@ if (apiKey !== undefined) {
   }
   const site = loadSite(prompt.site);
   try {
-    const grok = await callGrok({
+    const grok = await callLlm({
+      provider: provider(),
       apiKey,
+      model: provider() === 'openrouter' ? process.env.OPENROUTER_MODEL : undefined,
       prompt: prompt.prompt,
       siteId: site.id,
       builder: detectBuilder(site),
+      pages: site.posts.map((post) => post.slug),
     });
     grokLive = {
       id: prompt.id,
@@ -57,7 +72,10 @@ if (apiKey !== undefined) {
 const report = {
   fixturePassed: fixtureScores.filter((row) => row.passed).length,
   fixtureTotal: fixtureScores.length,
-  grokLive: grokLive ?? { skipped: true, reason: 'XAI_API_KEY not set' },
+  grokLive: grokLive ?? {
+    skipped: true,
+    reason: provider() === 'openrouter' ? 'OPENROUTER_API_KEY not set' : 'XAI_API_KEY not set',
+  },
   scores: fixtureScores,
 };
 

spikes/README.md

@@ -6,7 +6,7 @@ Throwaway fixtures and scripts are allowed. Product UI is not.
 
 | Spike | File | Status |
 |---|---|---|
-| Golden-task harness (R7) | [golden-task.md](./golden-task.md) | partial — live Grok pending key |
+| Golden-task harness (R7) | [golden-task.md](./golden-task.md) | done — live run scored via OpenRouter |
 | Builder detect (R6 / R13) | [builder-detect.md](./builder-detect.md) | done |
 | Pairing threat model (R9) | [pairing-threat-model.md](./pairing-threat-model.md) | done |
 | Large-site mirror timing (R4) | [mirror-timing.md](./mirror-timing.md) | done — synthetic 2GB |

spikes/golden-task.md

@@ -1,6 +1,6 @@
 # Spike: golden-task harness (R7)
 
-**Status:** partial — harness exists; live Grok run not scored (`XAI_API_KEY` unset)
+**Status:** done — live run scored via OpenRouter (`x-ai/grok-4.6`), `gb-01` passed
 
 ## Question
 
@@ -15,7 +15,7 @@ Can we score a model on WordPress tasks without vibes?
 
 ## Result
 
-Yes, if “score” means: apply a tool call to a fixture and assert the new heading/option. No, if it means we have a Grok quality number. This machine has no `XAI_API_KEY`, so the live call was skipped.
+Yes, if “score” means: apply a tool call to a fixture and assert the new heading/option. The harness is now live-scored through OpenRouter.
 
 ### What exists
 
@@ -24,22 +24,28 @@ Yes, if “score” means: apply a tool call to a fixture and assert the new hea
 | 20 prompts | `e2e/golden/prompts.json` |
 | Gutenberg dental site | `e2e/golden/sites/gutenberg-business/site.json` |
 | Elementor restaurant site | `e2e/golden/sites/elementor-restaurant/site.json` |
-| Apply + assert + Grok parser | `e2e/golden/src/` |
+| Apply + assert + LLM parser | `e2e/golden/src/` |
+| Provider client (grok + openrouter) | `e2e/golden/src/llm-client.ts` |
 | Scoreboard | `e2e/golden/runs/latest.json` |
 
 Two sites. Ten prompts each. Assertions are `preview_text`, `option`, or `screenshot`. Screenshot here means “the fixture page text must contain X” — not a PNG/SSIM check.
 
-`pnpm --filter @wursor/e2e golden` scored **20/20** fixture tool traces. Live Grok: skipped.
+`pnpm --filter @wursor/e2e golden` scored **20/20** fixture tool traces and **gb-01 passed live** (`x-ai/grok-4.6` via OpenRouter).
 
-`pnpm test:e2e` — 21 tests, including the scorer.
+`pnpm test:e2e` — 28 tests, including the scorer and the provider client.
 
-### How to score a real Grok run
+### How to score a real run
 
 ```bash
-XAI_API_KEY=… pnpm --filter @wursor/e2e golden
+# .env: LLM_PROVIDER=openrouter, OPENROUTER_API_KEY=…
+pnpm --filter @wursor/e2e golden
 ```
 
-That sends `gb-01` through `api.x.ai` and asserts the homepage heading.
+Sends `gb-01` through OpenRouter and asserts the homepage heading.
+
+### Finding
+
+The model needs the page slugs in context. The first live call returned `page: "home"` and failed; after passing `pages=homepage,about,…` in the prompt, `gb-01` passed. The real agent must ship page/site context with every tool-call prompt.
 
 ### Decision