@steipete/oracle 0.17.1 → 0.17.3

package/dist/docs-site/gemini.html

@@ -256,7 +256,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <li><strong>Set Environment Variable:</strong> Export the key as <code>GEMINI_API_KEY</code>.</li>
 <p>``<code>bash export GEMINI_API_KEY=&quot;your-google-api-key&quot; </code>``</p>
 <li><strong>Run Oracle:</strong> Use the <code>--model</code> (or <code>-m</code>) flag to select Gemini.</li>
-<p>``<code>bash oracle --engine api --model gemini --prompt &quot;Explain quantum entanglement&quot; </code>`<code> Use an explicit current model ID: </code>`<code>bash oracle --engine api --model gemini-3.5-flash --prompt &quot;...&quot; </code>`<code> Gemini 3.1 Pro is also available; Oracle dispatches it to Google&#39;s preview model id: </code>`<code>bash oracle --engine api --model gemini-3.1-pro --prompt &quot;...&quot; </code>`<code> For the lowest-cost current model: </code>`<code>bash oracle --engine api --model gemini-3.1-flash-lite --prompt &quot;...&quot; </code>``</p>
+<p>``<code>bash oracle --engine api --model gemini --prompt &quot;Explain quantum entanglement&quot; </code>`<code> You can also use the explicit model ID: </code>`<code>bash oracle --engine api --model gemini-3-pro --prompt &quot;...&quot; </code>`<code> Or the 3.1 alias, which Oracle dispatches to Google&#39;s preview model id: </code>`<code>bash oracle --engine api --model gemini-3.1-pro --prompt &quot;...&quot; </code>``</p>
 </ol>
 <h2 id="usage-gemini-web-cookies"><a class="anchor" href="#usage-gemini-web-cookies" aria-label="Anchor link">#</a>Usage (Gemini web / cookies)</h2>
 <p>Gemini web mode is a cookie-based client for <code>gemini.google.com</code>. It does <strong>not</strong> use <code>GEMINI_API_KEY</code> and does <strong>not</strong> drive ChatGPT.</p>
@@ -267,7 +267,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 </ul>
 <p>Examples:</p>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#768390"># Text run</span></span>
-<span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3.5-flash</span><span style="color:#6CB6FF"> --prompt</span><span style="color:#96D0FF"> "Say OK."</span></span>
+<span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3-pro</span><span style="color:#6CB6FF"> --prompt</span><span style="color:#96D0FF"> "Say OK."</span></span>
 <span class="line"></span>
 <span class="line"><span style="color:#768390"># Deep Think browser run (manual-login profile recommended on macOS)</span></span>
 <span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --browser-manual-login</span><span style="color:#F47067"> \</span></span>
@@ -275,30 +275,27 @@ body:not(.home) .doc>h1:first-child{display:none}
 <span class="line"><span style="color:#6CB6FF">  --prompt</span><span style="color:#96D0FF"> "Think carefully, then answer in one paragraph."</span></span>
 <span class="line"></span>
 <span class="line"><span style="color:#768390"># Generate an image (writes an output file)</span></span>
-<span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3.1-pro</span><span style="color:#F47067"> \</span></span>
+<span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3-pro</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --prompt</span><span style="color:#96D0FF"> "a cute robot holding a banana"</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --generate-image</span><span style="color:#96D0FF"> out.jpg</span><span style="color:#6CB6FF"> --aspect</span><span style="color:#96D0FF"> 1:1</span></span>
 <span class="line"></span>
 <span class="line"><span style="color:#768390"># Edit an image (input via --edit-image, output via --output)</span></span>
-<span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3.1-pro</span><span style="color:#F47067"> \</span></span>
+<span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3-pro</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --prompt</span><span style="color:#96D0FF"> "add sunglasses"</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --edit-image</span><span style="color:#96D0FF"> in.png</span><span style="color:#6CB6FF"> --output</span><span style="color:#96D0FF"> out.jpg</span></span></code></pre>
 <p>Notes:</p>
 <ul>
-<li>Current explicit IDs are <code>gemini-3.1-flash-lite</code>, <code>gemini-3.5-flash</code>, and <code>gemini-3.1-pro</code>.</li>
-<li>Legacy <code>gemini-3-pro</code>, <code>gemini-2.5-pro</code>, and <code>gemini-2.5-flash</code> browser names remain accepted and map to current Gemini web models.</li>
-<li>If your logged-in Gemini account can’t access the requested model, Oracle auto-falls back to Gemini 3.1 Flash-Lite and logs the fallback in verbose mode.</li>
+<li>If your logged-in Gemini account can’t access “Pro”, Oracle will auto-fallback to a supported model for web runs (and logs the fallback in verbose mode).</li>
 <li>This path runs fully in Node/TypeScript (no Python/venv dependency).</li>
 <li><code>--browser-model-strategy</code> only affects ChatGPT automation; Gemini web always uses the explicit Gemini model ID.</li>
 <li><code>gemini-3-deep-think</code> is browser-only for now. <code>--engine api</code> rejects it instead of silently falling back to regular Gemini Pro.</li>
-<li>Oracle intentionally does not expose generic <code>low</code> / <code>medium</code> / <code>high</code> Gemini aliases. Explicit IDs keep model choice, billing, and thinking-effort configuration distinct.</li>
 <li>If Chrome cookie extraction fails, the missing-cookie error now includes any cookie-reader warnings plus <code>--browser-manual-login</code> / <code>--browser-inline-cookies-file</code> guidance.</li>
 </ul>
 <h2 id="implementation-details"><a class="anchor" href="#implementation-details" aria-label="Anchor link">#</a>Implementation details</h2>
 <h3 id="gemini-api-adapter"><a class="anchor" href="#gemini-api-adapter" aria-label="Anchor link">#</a>Gemini API adapter</h3>
 <ul>
 <li><code>src/oracle/gemini.ts</code> — adapter using <code>@google/genai</code> that returns a <code>ClientLike</code>.</li>
-<li>Model IDs: <code>gemini-3.1-flash-lite</code> and <code>gemini-3.5-flash</code> use their stable API IDs; <code>gemini-3.1-pro</code> maps to <code>gemini-3.1-pro-preview</code>; legacy <code>gemini-3-pro</code> maps to <code>gemini-3-pro-preview</code>.</li>
+<li>Model IDs: <code>gemini-3-pro</code> maps to <code>gemini-3-pro-preview</code>; <code>gemini-3.1-pro</code> maps to <code>gemini-3.1-pro-preview</code>.</li>
 <li>Request mapping: <code>OracleRequestBody</code> → Gemini request; <code>web_search_preview</code> maps to Gemini search tooling.</li>
 <li>Response mapping: Gemini responses → <code>OracleResponse</code>.</li>
 <li>Streaming: wraps Gemini’s async iterator as <code>ResponseStreamLike</code>.</li>
@@ -307,7 +304,6 @@ body:not(.home) .doc>h1:first-child{display:none}
 </ul>
 <h3 id="gemini-web-client-cookie-based"><a class="anchor" href="#gemini-web-client-cookie-based" aria-label="Anchor link">#</a>Gemini web client (cookie-based)</h3>
 <ul>
-<li><code>src/gemini-web/models.ts</code> — centralizes current private web model headers, legacy aliases, and fallback selection.</li>
 <li><code>src/gemini-web/client.ts</code> — talks to <code>gemini.google.com</code> and downloads generated images via authenticated <code>gg-dl</code> redirects.</li>
 <li><code>src/gemini-web/executor.ts</code> — browser-engine executor for Gemini (loads Chrome cookies and runs the web client).</li>
 </ul>

package/dist/docs-site/index.html

@@ -4,19 +4,19 @@
   <meta charset="utf-8">
   <meta name="viewport" content="width=device-width, initial-scale=1">
   <title>oracle — Whisper your prompt to a mythical pro agent</title>
-  <meta name="description" content="Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.">
+  <meta name="description" content="Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.">
   <link rel="canonical" href="https://askoracle.sh/">
   <meta property="og:type" content="website">
   <meta property="og:site_name" content="oracle">
   <meta property="og:title" content="oracle — Whisper your prompt to a mythical pro agent">
-  <meta property="og:description" content="Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.">
+  <meta property="og:description" content="Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.">
   <meta property="og:url" content="https://askoracle.sh/">
   <meta property="og:image" content="https://askoracle.sh/social-card.png">
   <meta property="og:image:width" content="1200">
   <meta property="og:image:height" content="630">
   <meta name="twitter:card" content="summary_large_image">
   <meta name="twitter:title" content="oracle — Whisper your prompt to a mythical pro agent">
-  <meta name="twitter:description" content="Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.">
+  <meta name="twitter:description" content="Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.">
   <meta name="twitter:image" content="https://askoracle.sh/social-card.png">
   <link rel="icon" href="favicon.svg" type="image/svg+xml">
   <link rel="preconnect" href="https://fonts.googleapis.com">
@@ -236,7 +236,7 @@ body:not(.home) .doc>h1:first-child{display:none}
       <header class="home-hero">
         <p class="eyebrow">One CLI · Every Pro agent</p>
         <h1>Whisper your prompt to a mythical pro agent</h1>
-        <p class="lede">Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.</p>
+        <p class="lede">Oracle bundles your prompt and files so a mythical pro agent — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, and friends — can answer with real repository context. CLI, MCP, browser, and API in one tool.</p>
         <div class="home-cta">
           <a class="btn btn-primary" href="quickstart.html">Quickstart</a>
           <a class="btn btn-ghost" href="https://github.com/steipete/oracle" rel="noopener">GitHub</a>
@@ -272,7 +272,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <p><code>--render</code> emits the assembled markdown, sessions persist machine-readable metadata, and progress stays out of saved answers so pipes remain usable.</p>
 <h2 id="what-oracle-does"><a class="anchor" href="#what-oracle-does" aria-label="Anchor link">#</a>What Oracle does</h2>
 <ul>
-<li><strong>One CLI to a stable of pro agents.</strong> GPT-5.5 Pro (default), GPT-5.5, GPT-5.4 Pro, GPT-5.4, GPT-5.2 Pro, GPT-5.1 Pro, GPT-5.1 Codex, Gemini 3.1 Pro, Gemini 3.5 Flash, Gemini 3.1 Flash-Lite, Claude Sonnet 4.6, Claude Opus 4.1 — plus any OpenRouter id.</li>
+<li><strong>One CLI to a stable of pro agents.</strong> GPT-5.5 Pro (default), GPT-5.5, GPT-5.4 Pro, GPT-5.4, GPT-5.2 Pro, GPT-5.1 Pro, GPT-5.1 Codex, Gemini 3.1 Pro, Gemini 3 Pro, Claude Sonnet 4.6, Claude Opus 4.1 — plus any OpenRouter id.</li>
 <li><strong>Engines, plural.</strong> API mode for reliability, browser mode (Chrome over CDP) when you don&#39;t want to pay or want the Pro tier, <code>--render --copy</code> when neither is an option.</li>
 <li><strong>Multi-model in one run.</strong> Aggregate cost, token usage, and lineage across providers in a single command.</li>
 <li><strong>Recoverable panels.</strong> <code>doctor --providers</code>, <code>--preflight</code>, <code>--route</code>, and <code>--allow-partial</code> make provider/key failures clear without losing successful model output.</li>
@@ -285,13 +285,13 @@ body:not(.home) .doc>h1:first-child{display:none}
 <h2 id="pick-your-path"><a class="anchor" href="#pick-your-path" aria-label="Anchor link">#</a>Pick your path</h2>
 <ul>
 <li><strong>Trying it.</strong> <a href="install.html">Install</a> → <a href="quickstart.html">Quickstart</a>. Five minutes from <code>brew install</code> to your first answer.</li>
-<li><strong>Choosing a model.</strong> The <a href="mythical-pro-agents.html">Mythical Pro Agents</a> lineup covers when to reach for GPT-5.5 Pro vs. Gemini 3.1 Pro vs. Claude Opus, and what each costs.</li>
+<li><strong>Choosing a model.</strong> The <a href="mythical-pro-agents.html">Mythical Pro Agents</a> lineup covers when to reach for GPT-5.5 Pro vs. Gemini 3 Pro vs. Claude Opus, and what each costs.</li>
 <li><strong>Wiring up an agent.</strong> <a href="agents.html">Agents</a> covers Claude Code, Codex, Cursor, and the <code>oracle</code> skill. <a href="mcp.html">MCP</a> plugs Oracle into any MCP-aware client.</li>
 <li><strong>Driving ChatGPT without keys.</strong> <a href="browser-mode.html">Browser mode</a> walks through manual-login profiles, attach-running, remote browsers, and Deep Research.</li>
 <li><strong>Long Pro runs.</strong> <a href="sessions.html">Sessions</a> and the <a href="followup.html">followup</a> flow handle background runs, reattach, and lineage.</li>
 </ul>
 <h2 id="why-mythical-pro-agents"><a class="anchor" href="#why-mythical-pro-agents" aria-label="Anchor link">#</a>Why &quot;mythical pro agents&quot;?</h2>
-<p>The frontier models marked &quot;Pro&quot; — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, OpenAI Deep Research — are slow, expensive, and gated behind ChatGPT Plus / Pro tiers, separate APIs, or per-token bills that scale fast. Oracle is the single entry point: one config, one session store, one set of flags. Bundle the right files, ask the right model, get a second opinion without remembering which provider charges for what.</p>
+<p>The frontier models marked &quot;Pro&quot; — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, OpenAI Deep Research — are slow, expensive, and gated behind ChatGPT Plus / Pro tiers, separate APIs, or per-token bills that scale fast. Oracle is the single entry point: one config, one session store, one set of flags. Bundle the right files, ask the right model, get a second opinion without remembering which provider charges for what.</p>
 <h2 id="project"><a class="anchor" href="#project" aria-label="Anchor link">#</a>Project</h2>
 <p>Active development under MIT. The <a href="https://github.com/steipete/oracle/blob/main/CHANGELOG.md">changelog</a> tracks recent releases. Source on <a href="https://github.com/steipete/oracle">GitHub</a>. Not affiliated with OpenAI, Google, or Anthropic.</p></article>
         

package/dist/docs-site/install.html

@@ -260,7 +260,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <p><code>npx</code> is fine for CI, ad-hoc scripts, or when you don&#39;t want a global binary on the box. Cache the package in CI by pinning the version (<code>@steipete/oracle@0.12.1</code>) so you don&#39;t re-download on every job.</p>
 <h2 id="api-keys-optional"><a class="anchor" href="#api-keys-optional" aria-label="Anchor link">#</a>API keys (optional)</h2>
 <p>API mode is opt-in and reads keys from the environment. Set whichever providers you&#39;ll use:</p>
-<table><thead><tr><th>Provider</th><th>Env var</th><th>Models</th></tr></thead><tbody><tr><td>OpenAI</td><td><code>OPENAI_API_KEY</code></td><td>GPT-5.x, GPT-5.x Pro, GPT-5.1 Codex</td></tr><tr><td>Azure OpenAI</td><td><code>AZURE_OPENAI_API_KEY</code>, <code>AZURE_OPENAI_ENDPOINT</code>, <code>..._DEPLOYMENT</code></td><td>Same models, hosted on Azure</td></tr><tr><td>Google</td><td><code>GEMINI_API_KEY</code></td><td>Gemini 3.1 Pro, Gemini 3.5 Flash, Gemini 3.1 Flash-Lite</td></tr><tr><td>Anthropic</td><td><code>ANTHROPIC_API_KEY</code></td><td>Claude Sonnet 4.6, Claude Opus 4.1</td></tr><tr><td>OpenRouter</td><td><code>OPENROUTER_API_KEY</code></td><td>Any OpenRouter id (e.g. <code>minimax/minimax-m2</code>)</td></tr></tbody></table>
+<table><thead><tr><th>Provider</th><th>Env var</th><th>Models</th></tr></thead><tbody><tr><td>OpenAI</td><td><code>OPENAI_API_KEY</code></td><td>GPT-5.x, GPT-5.x Pro, GPT-5.1 Codex</td></tr><tr><td>Azure OpenAI</td><td><code>AZURE_OPENAI_API_KEY</code>, <code>AZURE_OPENAI_ENDPOINT</code>, <code>..._DEPLOYMENT</code></td><td>Same models, hosted on Azure</td></tr><tr><td>Google</td><td><code>GEMINI_API_KEY</code></td><td>Gemini 3.1 Pro (API-only), Gemini 3 Pro</td></tr><tr><td>Anthropic</td><td><code>ANTHROPIC_API_KEY</code></td><td>Claude Sonnet 4.6, Claude Opus 4.1</td></tr><tr><td>OpenRouter</td><td><code>OPENROUTER_API_KEY</code></td><td>Any OpenRouter id (e.g. <code>minimax/minimax-m2</code>)</td></tr></tbody></table>
 <p>If no key is set, Oracle defaults to <strong>browser mode</strong> and drives ChatGPT directly — see <a href="browser-mode.html">Browser Mode</a> for the manual-login flow.</p>
 <h2 id="where-oracle-stores-state"><a class="anchor" href="#where-oracle-stores-state" aria-label="Anchor link">#</a>Where Oracle stores state</h2>
 <table><thead><tr><th>Path</th><th>Contents</th></tr></thead><tbody><tr><td><code>~/.oracle/config.json</code></td><td>Defaults (JSON5). See <a href="configuration.html">Configuration</a>.</td></tr><tr><td><code>~/.oracle/sessions/&lt;id&gt;/</code></td><td>Run logs, bundles, transcripts, generated artifacts</td></tr><tr><td><code>~/.oracle/cookies.json</code></td><td>(Optional) inline ChatGPT cookies for browser mode</td></tr></tbody></table>

package/dist/docs-site/manual-tests.html

@@ -433,7 +433,6 @@ body:not(.home) .doc>h1:first-child{display:none}
 <li><strong>GPT-5.5 simple prompt</strong></li>
 <p><code>pnpm run oracle -- --engine browser --model gpt-5.5 --prompt &quot;List two reasons Markdown is handy&quot;</code> Confirm the answer arrives (and only once) even if it takes ~2–3 minutes.</p>
 </ol>
-<p>2b. <strong>GPT-5.5 Instant smoke</strong> <code>pnpm run oracle -- --engine browser --model gpt-5.5-instant --prompt &quot;Give me two short markdown bullet points about tables&quot;</code> Expect a near-instant response (no Thinking spinner) and confirm the composer pill shows the &quot;Instant&quot; row, not &quot;Thinking 5.5&quot; or &quot;Pro&quot;. Run after any change to the 5.5 picker tokens.</p>
 <ol>
 <li><strong>GPT-5.5 + attachment</strong></li>
 <p>Prepare <code>/tmp/browser-md.txt</code> with a short note, then run <code>pnpm run oracle -- --engine browser --model gpt-5.5 --prompt &quot;Summarize the key idea from the attached note&quot; --file /tmp/browser-md.txt</code> Ensure upload logs show “Attachment queued” and the answer references the file contents explicitly.</p>
@@ -525,7 +524,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <li>Export a real key and explicitly opt in (default runs stay fast):</li>
 <p>``<code>bash export OPENAI_API_KEY=sk-... export ORACLE_LIVE_TEST=1 pnpm vitest run tests/live/openai-live.test.ts </code>``</p>
 <li>The first two tests target the standard GPT-5 (<code>gpt-5.1</code> / <code>gpt-5.2</code>) foreground</li>
-<p>streaming paths. The later background tests send <code>gpt-5.5-pro</code> and GPT-5.6 Sol with Pro mode and max reasoning effort prompts and expect the CLI to stay in background mode until OpenAI finishes (up to 30 minutes).</p>
+<p>streaming paths. The later background tests send <code>gpt-5.5-pro</code> and <code>gpt-5.2-pro</code> prompts and expect the CLI to stay in background mode until OpenAI finishes (up to 30 minutes).</p>
 <li>Watch the console for <code>Reconnected to OpenAI background response...</code> if</li>
 <p>you&#39;re debugging transport flakiness; the test will fail if the response status isn&#39;t <code>completed</code> or if the text doesn&#39;t contain the hard-coded smoke strings.</p>
 </ol>

package/dist/docs-site/mcp.html

@@ -249,43 +249,20 @@ body:not(.home) .doc>h1:first-child{display:none}
 <h2 id="let-them-fight"><a class="anchor" href="#let-them-fight" aria-label="Anchor link">#</a>Let Them Fight</h2>
 <p>Claude Code can call <code>oracle-mcp</code> and ask a subscription-backed ChatGPT browser session for a second opinion. Use the <code>chatgpt-pro-heavy</code> preset when you want a compact MCP request that targets ChatGPT browser mode, the current Pro picker alias, and Pro Extended thinking time. The preset is intentionally boring at the API layer: it is a shortcut for existing browser-mode fields, not a new model id.</p>
 <h2 id="tools"><a class="anchor" href="#tools" aria-label="Anchor link">#</a>Tools</h2>
-<h3 id="chatgpt-image"><a class="anchor" href="#chatgpt-image" aria-label="Anchor link">#</a><code>chatgpt_image</code></h3>
-<ul>
-<li>Inputs: <code>prompt</code> (required), <code>files?: string[]</code> for reference images/assets, <code>outputPath?: string</code>, <code>aspectRatio?: string</code>, <code>model?: string</code>, plus browser controls such as <code>browserThinkingTime</code>, <code>browserModelLabel</code>, <code>browserModelStrategy</code>, <code>browserArchive</code>, <code>browserKeepBrowser</code>, and <code>dryRun</code>.</li>
-<li>Behavior: convenience wrapper for ChatGPT browser image generation. It forces <code>engine:&quot;browser&quot;</code>, sets <code>generateImage</code> for the existing image-aware wait/download path, and defaults <code>browserAttachments:&quot;always&quot;</code> when files are provided so reference images are uploaded instead of pasted.</li>
-<li>Output: returns the normal session metadata plus <code>requestedOutputPath</code> and <code>structuredContent.images[]</code> with saved local paths, MIME type, size, dimensions, and ChatGPT file id when available. Signed source/download URLs are not returned. If <code>outputPath</code> is omitted, Oracle picks a unique file under <code>ORACLE_HOME_DIR/generated/</code>.</li>
-<li>Output path safety: agent-supplied <code>outputPath</code> must resolve under <code>ORACLE_HOME_DIR/generated</code> by default; traversal and symlink escapes are rejected. This keeps MCP writes away from Oracle config, session metadata, and browser profile state. Set <code>ORACLE_MCP_ALLOW_EXTERNAL_OUTPUT=1</code> to allow writing elsewhere as an explicit operator decision. Omit <code>outputPath</code> to use the safe default.</li>
-<li>Local browser only: image output is unsupported when a remote browser service is configured (<code>ORACLE_REMOTE_HOST</code>); the image would be written on the remote host and not transferred back, so <code>chatgpt_image</code>/<code>consult</code> image runs fail closed with a clear error rather than returning empty <code>structuredContent.images</code>. Run on the local browser to generate images.</li>
-</ul>
-<pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-json"><span class="line"><span style="color:#ADBAC7">{</span></span>
-<span class="line"><span style="color:#8DDB8C">  "prompt"</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"Create a 9:16 App Store screenshot background for a focus timer."</span><span style="color:#ADBAC7">,</span></span>
-<span class="line"><span style="color:#8DDB8C">  "files"</span><span style="color:#ADBAC7">: [</span><span style="color:#96D0FF">"./reference-screen.png"</span><span style="color:#ADBAC7">],</span></span>
-<span class="line"><span style="color:#8DDB8C">  "aspectRatio"</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"9:16"</span></span>
-<span class="line"><span style="color:#ADBAC7">}</span></span></code></pre>
 <h3 id="consult"><a class="anchor" href="#consult" aria-label="Anchor link">#</a><code>consult</code></h3>
 <ul>
-<li>Inputs: <code>prompt</code> (required), <code>files?: string[]</code> (globs), <code>model?: string</code> (defaults to CLI), <code>engine?: &quot;api&quot; | &quot;browser&quot;</code> (optional; Oracle follows CLI defaults: <code>ORACLE_ENGINE</code> and the effective config first, then API when <code>OPENAI_API_KEY</code> is set, otherwise browser), <code>slug?: string</code>.</li>
+<li>Inputs: <code>prompt</code> (required), <code>files?: string[]</code> (globs), <code>model?: string</code> (defaults to CLI), <code>engine?: &quot;api&quot; | &quot;browser&quot;</code> (optional; Oracle follows CLI defaults: config/<code>ORACLE_ENGINE</code> first, then API when <code>OPENAI_API_KEY</code> is set, otherwise browser), <code>slug?: string</code>.</li>
 <li>Presets: <code>preset?: &quot;chatgpt-pro-heavy&quot;</code> applies browser mode + current Pro model alias + extended thinking, unless the request overrides those fields.</li>
-<li>Browser-only extras: <code>browserAttachments?: &quot;auto&quot;|&quot;never&quot;|&quot;always&quot;</code>, <code>browserBundleFiles?: boolean</code>, <code>browserBundleFormat?: &quot;auto&quot;|&quot;text&quot;|&quot;zip&quot;</code>, <code>browserThinkingTime?: &quot;light&quot;|&quot;standard&quot;|&quot;extended&quot;|&quot;extra-high&quot;|&quot;heavy&quot;</code>, <code>browserResearchMode?: &quot;deep&quot;</code>, <code>browserFollowUps?: string[]</code>, <code>browserArchive?: &quot;auto&quot;|&quot;always&quot;|&quot;never&quot;</code>, <code>browserKeepBrowser?: boolean</code>, <code>browserModelLabel?: string</code>, <code>browserModelStrategy?: &quot;select&quot;|&quot;current&quot;|&quot;ignore&quot;</code>, <code>generateImage?: string</code>, <code>outputPath?: string</code>.</li>
+<li>Browser-only extras: <code>browserAttachments?: &quot;auto&quot;|&quot;never&quot;|&quot;always&quot;</code>, <code>browserBundleFiles?: boolean</code>, <code>browserBundleFormat?: &quot;text&quot;|&quot;zip&quot;</code>, <code>browserThinkingTime?: &quot;light&quot;|&quot;standard&quot;|&quot;extended&quot;|&quot;heavy&quot;</code>, <code>browserResearchMode?: &quot;deep&quot;</code>, <code>browserFollowUps?: string[]</code>, <code>browserArchive?: &quot;auto&quot;|&quot;always&quot;|&quot;never&quot;</code>, <code>browserKeepBrowser?: boolean</code>, <code>browserModelLabel?: string</code>, <code>browserModelStrategy?: &quot;select&quot;|&quot;current&quot;|&quot;ignore&quot;</code>.</li>
 <li>Dry runs: set <code>dryRun: true</code> to preview the resolved request without creating a session or touching the browser.</li>
 <li>Behavior: starts a session, runs it with the chosen engine, returns final output + metadata. Background/foreground follows the CLI (e.g., GPT‑5 Pro detaches by default). If API mode fails because <code>OPENAI_API_KEY</code> is missing and you have ChatGPT Pro, retry with <code>engine: &quot;browser&quot;</code> or <code>preset: &quot;chatgpt-pro-heavy&quot;</code> to use your signed-in ChatGPT session instead of an API key.</li>
 <li>Logging: emits MCP logs (<code>info</code> per line, <code>debug</code> for streamed chunks with byte sizes). If browser prerequisites are missing, returns an error payload instead of running.</li>
-<li>Research mode: set <code>browserResearchMode:&quot;deep&quot;</code> for broad public-web research and cited reports. Use normal browser runs with <code>gpt-5.5-pro</code> + <code>browserThinkingTime:&quot;extended&quot;</code> for Pro Extended code review, <code>gpt-5.6-sol</code> + <code>browserThinkingTime:&quot;extra-high&quot;</code> for Extra High, or <code>browserThinkingTime:&quot;heavy&quot;</code> when you explicitly want Pro.</li>
+<li>Research mode: set <code>browserResearchMode:&quot;deep&quot;</code> for broad public-web research and cited reports. Use normal browser runs with <code>gpt-5.5-pro</code> + <code>browserThinkingTime:&quot;extended&quot;</code> for Pro Extended code review, or <code>gpt-5.5</code> + <code>browserThinkingTime:&quot;heavy&quot;</code> when you explicitly want Thinking Heavy.</li>
 <li>Multi-turn consults: set <code>browserFollowUps:[&quot;Challenge your recommendation&quot;, &quot;Give the final decision&quot;]</code> to keep one ChatGPT browser conversation open and ask sequential follow-up prompts. Use one-shot calls for narrow bugs and exact file-set reviews; use multi-turn for ambiguous architecture/product decisions where a challenge pass and final recommendation are useful; use Deep Research for broad public-web work with citations. Oracle never invents follow-ups automatically.</li>
 <li>Archiving: set <code>browserArchive:&quot;auto&quot;|&quot;always&quot;|&quot;never&quot;</code> to control ChatGPT conversation cleanup. <code>auto</code> archives only successful browser one-shots after local artifacts are saved, and skips project, Deep Research, multi-turn, failed, and incomplete sessions.</li>
-<li>ChatGPT image generation: set <code>engine:&quot;browser&quot;</code> and <code>generateImage</code> to a path under <code>ORACLE_HOME_DIR/generated</code> to use the same image-aware wait/download path as CLI <code>--generate-image</code>. Saved files are returned in <code>structuredContent.images</code> and recorded as session artifacts; multiple images save as numbered siblings. Agent-supplied <code>generateImage</code> / <code>outputPath</code> are constrained to that generated-output directory by default (set <code>ORACLE_MCP_ALLOW_EXTERNAL_OUTPUT=1</code> to allow external paths).</li>
 </ul>
 <h4 id="long-browser-consults-from-agents"><a class="anchor" href="#long-browser-consults-from-agents" aria-label="Anchor link">#</a>Long browser consults from agents</h4>
 <p>Browser-backed GPT-5.5 Pro consults can legitimately run for many minutes. Some MCP clients show little progress while a tool call is active, so agents should treat a long Oracle call as a running browser job, not as a failed step. Start with <code>dryRun:true</code> when configuring a new agent, prefer <code>preset:&quot;chatgpt-pro-heavy&quot;</code> or <code>engine:&quot;browser&quot;</code> explicitly, and use the shared session store (<code>sessions</code>, <code>oracle status</code>, or <code>oracle session &lt;id&gt;</code>) before retrying a prompt. If the browser control plan says Oracle will launch visible Chrome, use attach/remote Chrome when the operator is actively using the computer.</p>
-<h4 id="chatgpt-images-from-agents"><a class="anchor" href="#chatgpt-images-from-agents" aria-label="Anchor link">#</a>ChatGPT images from agents</h4>
-<p>For generated images, pass an explicit <code>generateImage</code> path. That opt-in is important because it switches the browser wait loop to watch for ChatGPT image artifacts instead of only assistant text. The path must resolve under <code>ORACLE_HOME_DIR/generated</code> unless <code>ORACLE_MCP_ALLOW_EXTERNAL_OUTPUT=1</code> is set.</p>
-<pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-json"><span class="line"><span style="color:#ADBAC7">{</span></span>
-<span class="line"><span style="color:#8DDB8C">  "engine"</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"browser"</span><span style="color:#ADBAC7">,</span></span>
-<span class="line"><span style="color:#8DDB8C">  "model"</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"gpt-5.5-pro"</span><span style="color:#ADBAC7">,</span></span>
-<span class="line"><span style="color:#8DDB8C">  "prompt"</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"Create a 9:16 App Store screenshot background for a focus timer."</span><span style="color:#ADBAC7">,</span></span>
-<span class="line"><span style="color:#8DDB8C">  "generateImage"</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"${ORACLE_HOME_DIR}/generated/focus-timer-bg.png"</span></span>
-<span class="line"><span style="color:#ADBAC7">}</span></span></code></pre>
-<p>The MCP response includes <code>structuredContent.images[]</code> with the saved file path, MIME type, size, dimensions, and ChatGPT file id when available. Signed source/download URLs remain internal.</p>
 <h3 id="sessions"><a class="anchor" href="#sessions" aria-label="Anchor link">#</a><code>sessions</code></h3>
 <ul>
 <li>Inputs: <code>{id?, hours?, limit?, includeAll?, detail?}</code> mirroring <code>oracle status</code> / <code>oracle session</code>.</li>
@@ -323,9 +300,9 @@ body:not(.home) .doc>h1:first-child{display:none}
 <li>Claude Code: <code>oracle bridge claude-config</code></li>
 <li>Claude Code with local macOS Chrome: <code>oracle bridge claude-config --local-browser &gt; .mcp.json</code></li>
 <li>Tools and resources operate on the same session store as <code>oracle status|session</code>.</li>
-<li>Defaults (model/engine/etc.) come from the effective Oracle CLI config; see <code>docs/configuration.md</code>, <code>~/.oracle/config.json</code>, and project <code>.oracle/config.json</code> files.</li>
+<li>Defaults (model/engine/etc.) come from your Oracle CLI config; see <code>docs/configuration.md</code> or <code>~/.oracle/config.json</code>.</li>
 </ul><nav class="page-nav" aria-label="Pager"><a class="page-nav-prev" href="agents.html"><small>Previous</small><span>Coding Agents</span></a><a class="page-nav-next" href="sessions.html"><small>Next</small><span>Sessions</span></a></nav></article>
-        <nav class="toc" aria-label="On this page"><h2>On this page</h2><a class="toc-l2" href="#let-them-fight">Let Them Fight</a><a class="toc-l2" href="#tools">Tools</a><a class="toc-l3" href="#chatgpt-image">chatgpt_image</a><a class="toc-l3" href="#consult">consult</a><a class="toc-l3" href="#sessions">sessions</a><a class="toc-l3" href="#project-sources">project_sources</a><a class="toc-l2" href="#resources">Resources</a><a class="toc-l2" href="#background-detach-behavior">Background / detach behavior</a><a class="toc-l2" href="#launching-usage">Launching &amp;amp; usage</a></nav>
+        <nav class="toc" aria-label="On this page"><h2>On this page</h2><a class="toc-l2" href="#let-them-fight">Let Them Fight</a><a class="toc-l2" href="#tools">Tools</a><a class="toc-l3" href="#consult">consult</a><a class="toc-l3" href="#sessions">sessions</a><a class="toc-l3" href="#project-sources">project_sources</a><a class="toc-l2" href="#resources">Resources</a><a class="toc-l2" href="#background-detach-behavior">Background / detach behavior</a><a class="toc-l2" href="#launching-usage">Launching &amp;amp; usage</a></nav>
       </div>
     </main>
   </div>

package/dist/docs-site/mythical-pro-agents.html

@@ -4,19 +4,19 @@
   <meta charset="utf-8">
   <meta name="viewport" content="width=device-width, initial-scale=1">
   <title>Mythical Pro Agents — oracle</title>
-  <meta name="description" content="The frontier &#39;Pro&#39; model lineup Oracle speaks to — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, Deep Research — and when to reach for each.">
+  <meta name="description" content="The frontier &#39;Pro&#39; model lineup Oracle speaks to — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, Deep Research — and when to reach for each.">
   <link rel="canonical" href="https://askoracle.sh/mythical-pro-agents.html">
   <meta property="og:type" content="website">
   <meta property="og:site_name" content="oracle">
   <meta property="og:title" content="Mythical Pro Agents — oracle">
-  <meta property="og:description" content="The frontier &#39;Pro&#39; model lineup Oracle speaks to — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, Deep Research — and when to reach for each.">
+  <meta property="og:description" content="The frontier &#39;Pro&#39; model lineup Oracle speaks to — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, Deep Research — and when to reach for each.">
   <meta property="og:url" content="https://askoracle.sh/mythical-pro-agents.html">
   <meta property="og:image" content="https://askoracle.sh/social-card.png">
   <meta property="og:image:width" content="1200">
   <meta property="og:image:height" content="630">
   <meta name="twitter:card" content="summary_large_image">
   <meta name="twitter:title" content="Mythical Pro Agents — oracle">
-  <meta name="twitter:description" content="The frontier &#39;Pro&#39; model lineup Oracle speaks to — GPT-5.5 Pro, Gemini 3.1 Pro, Claude Opus, Deep Research — and when to reach for each.">
+  <meta name="twitter:description" content="The frontier &#39;Pro&#39; model lineup Oracle speaks to — GPT-5.5 Pro, Gemini 3 Pro, Claude Opus, Deep Research — and when to reach for each.">
   <meta name="twitter:image" content="https://askoracle.sh/social-card.png">
   <link rel="icon" href="favicon.svg" type="image/svg+xml">
   <link rel="preconnect" href="https://fonts.googleapis.com">
@@ -246,7 +246,7 @@ body:not(.home) .doc>h1:first-child{display:none}
       <div class="doc-grid">
         <article class="doc"><p>The headline frontier models — the ones marked <strong>Pro</strong> — are slow, expensive, and gated behind separate consumer subscriptions or per-token bills. Oracle is the single CLI that talks to all of them with the same flags, the same session store, and the same bundling rules.</p>
 <h2 id="the-lineup"><a class="anchor" href="#the-lineup" aria-label="Anchor link">#</a>The lineup</h2>
-<table><thead><tr><th>Model</th><th>Engine</th><th>API id</th><th>Browser model picker</th><th>Speciality</th></tr></thead><tbody><tr><td>GPT-5.5 Pro</td><td>API or browser</td><td><code>gpt-5.5-pro</code> (default)</td><td>&quot;GPT-5.5 Pro&quot; / &quot;5.5 Pro&quot;</td><td>Long-form code review, planning</td></tr><tr><td>GPT-5.5</td><td>API or browser</td><td><code>gpt-5.5</code></td><td>&quot;GPT-5.5&quot;</td><td>Fast everyday consults</td></tr><tr><td>GPT-5.4 Pro</td><td>API or browser</td><td><code>gpt-5.4-pro</code></td><td>&quot;5.4 Pro&quot;</td><td>Mature Pro workflow</td></tr><tr><td>GPT-5.4</td><td>API or browser</td><td><code>gpt-5.4</code></td><td>&quot;GPT-5.4&quot;</td><td>Mid-tier general</td></tr><tr><td>GPT-5.2 Pro</td><td>API or browser</td><td><code>gpt-5.2-pro</code></td><td>&quot;5.2 Pro&quot;</td><td>Heavy reasoning</td></tr><tr><td>GPT-5.2</td><td>API or browser</td><td><code>gpt-5.2</code></td><td>&quot;GPT-5.2&quot;</td><td>Mid-tier</td></tr><tr><td>GPT-5.1 Pro</td><td>API or browser</td><td><code>gpt-5.1-pro</code></td><td>&quot;5.1 Pro&quot;</td><td>Long context</td></tr><tr><td>GPT-5.1</td><td>API or browser</td><td><code>gpt-5.1</code></td><td>&quot;GPT-5.1&quot;</td><td>General</td></tr><tr><td>GPT-5.1 Codex</td><td>API only</td><td><code>gpt-5.1-codex</code></td><td>—</td><td>Code generation</td></tr><tr><td>Gemini 3.1 Pro</td><td>API or browser</td><td><code>gemini-3.1-pro</code></td><td>&quot;3.1 Pro&quot;</td><td>Long-context multimodal, image gen, YouTube</td></tr><tr><td>Gemini 3.5 Flash</td><td>API or browser</td><td><code>gemini-3.5-flash</code></td><td>&quot;3.5 Flash&quot;</td><td>Fast all-around Gemini work</td></tr><tr><td>Gemini 3.1 Flash-Lite</td><td>API or browser</td><td><code>gemini-3.1-flash-lite</code></td><td>&quot;3.1 Flash-Lite&quot;</td><td>Lowest-cost, fastest Gemini work</td></tr><tr><td>Claude Opus 4.1</td><td>API only</td><td><code>claude-4.1-opus</code></td><td>—</td><td>Deepest single-shot reasoning</td></tr><tr><td>Claude Sonnet 4.6</td><td>API only</td><td><code>claude-4.6-sonnet</code></td><td>—</td><td>Fast Claude</td></tr></tbody></table>
+<table><thead><tr><th>Model</th><th>Engine</th><th>API id</th><th>Browser model picker</th><th>Speciality</th></tr></thead><tbody><tr><td>GPT-5.5 Pro</td><td>API or browser</td><td><code>gpt-5.5-pro</code> (default)</td><td>&quot;GPT-5.5 Pro&quot; / &quot;5.5 Pro&quot;</td><td>Long-form code review, planning</td></tr><tr><td>GPT-5.5</td><td>API or browser</td><td><code>gpt-5.5</code></td><td>&quot;GPT-5.5&quot;</td><td>Fast everyday consults</td></tr><tr><td>GPT-5.4 Pro</td><td>API or browser</td><td><code>gpt-5.4-pro</code></td><td>&quot;5.4 Pro&quot;</td><td>Mature Pro workflow</td></tr><tr><td>GPT-5.4</td><td>API or browser</td><td><code>gpt-5.4</code></td><td>&quot;GPT-5.4&quot;</td><td>Mid-tier general</td></tr><tr><td>GPT-5.2 Pro</td><td>API or browser</td><td><code>gpt-5.2-pro</code></td><td>&quot;5.2 Pro&quot;</td><td>Heavy reasoning</td></tr><tr><td>GPT-5.2</td><td>API or browser</td><td><code>gpt-5.2</code></td><td>&quot;GPT-5.2&quot;</td><td>Mid-tier</td></tr><tr><td>GPT-5.1 Pro</td><td>API or browser</td><td><code>gpt-5.1-pro</code></td><td>&quot;5.1 Pro&quot;</td><td>Long context</td></tr><tr><td>GPT-5.1</td><td>API or browser</td><td><code>gpt-5.1</code></td><td>&quot;GPT-5.1&quot;</td><td>General</td></tr><tr><td>GPT-5.1 Codex</td><td>API only</td><td><code>gpt-5.1-codex</code></td><td>—</td><td>Code generation</td></tr><tr><td>Gemini 3.1 Pro</td><td>API only</td><td><code>gemini-3.1-pro</code></td><td>—</td><td>Long-context multimodal</td></tr><tr><td>Gemini 3 Pro</td><td>API or browser</td><td><code>gemini-3-pro</code></td><td>(auto on <code>gemini.google.com</code>)</td><td>Multimodal, image gen, YouTube, free in browser</td></tr><tr><td>Claude Opus 4.1</td><td>API only</td><td><code>claude-4.1-opus</code></td><td>—</td><td>Deepest single-shot reasoning</td></tr><tr><td>Claude Sonnet 4.6</td><td>API only</td><td><code>claude-4.6-sonnet</code></td><td>—</td><td>Fast Claude</td></tr></tbody></table>
 <p>Plus any <strong>OpenRouter</strong> id — e.g. <code>minimax/minimax-m2</code>, <code>openai/gpt-4o-mini</code>, <code>qwen/qwen-2.5-coder-32b-instruct</code> — when you set <code>OPENROUTER_API_KEY</code>.</p>
 <h2 id="when-to-reach-for-which"><a class="anchor" href="#when-to-reach-for-which" aria-label="Anchor link">#</a>When to reach for which</h2>
 <h3 id="gpt-5-5-pro-default"><a class="anchor" href="#gpt-5-5-pro-default" aria-label="Anchor link">#</a>GPT-5.5 Pro (default)</h3>
@@ -254,9 +254,9 @@ body:not(.home) .doc>h1:first-child{display:none}
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gpt-5.5-pro</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  -p</span><span style="color:#96D0FF"> "Plan the auth migration end-to-end"</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --file</span><span style="color:#96D0FF"> "src/auth/**"</span><span style="color:#6CB6FF"> --file</span><span style="color:#96D0FF"> "docs/auth.md"</span></span></code></pre>
-<h3 id="gemini-3-1-pro"><a class="anchor" href="#gemini-3-1-pro" aria-label="Anchor link">#</a>Gemini 3.1 Pro</h3>
+<h3 id="gemini-3-pro"><a class="anchor" href="#gemini-3-pro" aria-label="Anchor link">#</a>Gemini 3 Pro</h3>
 <p>Free in browser mode if you&#39;re signed into <code>gemini.google.com</code> in Chrome. Also the path for <strong>YouTube transcript analysis</strong> (<code>--youtube</code>) and <strong>image generation/edit</strong> (<code>--generate-image</code>, <code>--edit-image</code>).</p>
-<pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3.1-pro</span><span style="color:#F47067"> \</span></span>
+<pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gemini-3-pro</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --prompt</span><span style="color:#96D0FF"> "a minimalist eye-of-providence logo, vector"</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --generate-image</span><span style="color:#96D0FF"> out.png</span><span style="color:#6CB6FF"> --aspect</span><span style="color:#96D0FF"> 1:1</span></span></code></pre>
 <h3 id="claude-opus-4-1"><a class="anchor" href="#claude-opus-4-1" aria-label="Anchor link">#</a>Claude Opus 4.1</h3>
@@ -278,7 +278,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <p>The captured report lands at <code>~/.oracle/sessions/&lt;id&gt;/artifacts/deep-research-report.md</code>. Deep Research can run for an hour or more — Oracle stores the session and reattaches.</p>
 <h2 id="thinking-time"><a class="anchor" href="#thinking-time" aria-label="Anchor link">#</a>Thinking time</h2>
 <p>Pro / Thinking models in browser mode accept a <code>--browser-thinking-time</code> knob:</p>
-<table><thead><tr><th>Level</th><th>What it maps to in ChatGPT</th></tr></thead><tbody><tr><td><code>light</code></td><td>Instant / Quick</td></tr><tr><td><code>standard</code></td><td>Medium / Standard</td></tr><tr><td><code>extended</code></td><td>High / Pro Extended / Thinking Extended</td></tr><tr><td><code>extra-high</code></td><td>Extra High (GPT-5.6 Sol non-Pro max)</td></tr><tr><td><code>heavy</code></td><td>Matching Heavy row; otherwise keep the current tier</td></tr></tbody></table>
+<table><thead><tr><th>Level</th><th>What it maps to in ChatGPT</th></tr></thead><tbody><tr><td><code>light</code></td><td>Quick</td></tr><tr><td><code>standard</code></td><td>Default</td></tr><tr><td><code>extended</code></td><td>Pro Extended / Thinking Extended</td></tr><tr><td><code>heavy</code></td><td>Heavy thinking</td></tr></tbody></table>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gpt-5.5-pro</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --browser-thinking-time</span><span style="color:#96D0FF"> extended</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  -p</span><span style="color:#96D0FF"> "Refactor this hot path"</span><span style="color:#6CB6FF"> --file</span><span style="color:#96D0FF"> "src/render/**"</span></span></code></pre>
@@ -286,15 +286,15 @@ body:not(.home) .doc>h1:first-child{display:none}
 <ul>
 <li><strong>GPT-5.x Pro</strong> (API): tokens cost meaningfully more than non-Pro. Watch the run summary.</li>
 <li><strong>GPT-5.x Pro</strong> (browser): &quot;free&quot; with ChatGPT Pro / Plus subscription, but slow.</li>
-<li><strong>Gemini 3.1 Pro / 3.5 Flash / 3.1 Flash-Lite</strong> (browser): available through a signed-in Google account, subject to account access.</li>
+<li><strong>Gemini 3 Pro</strong> (browser): free with a Google account.</li>
 <li><strong>Claude Opus 4.1</strong>: per-token API only.</li>
 <li><strong>OpenRouter ids</strong>: pricing varies wildly per provider; always preview with <code>--dry-run summary</code>.</li>
 </ul>
 <p><code>--files-report</code> plus <code>--dry-run summary</code> is the right reflex before any Pro run on a large bundle. Token counts ≠ dollars, but they&#39;re a close-enough proxy.</p>
 <h2 id="engine-compatibility"><a class="anchor" href="#engine-compatibility" aria-label="Anchor link">#</a>Engine compatibility</h2>
-<table><thead><tr><th>Capability</th><th>API</th><th>Browser (ChatGPT)</th><th>Browser (Gemini)</th></tr></thead><tbody><tr><td>GPT-5.x family</td><td>✅</td><td>✅</td><td>—</td></tr><tr><td>Gemini 3.1 Pro / Flash</td><td>✅</td><td>—</td><td>✅</td></tr><tr><td>Claude Sonnet / Opus</td><td>✅</td><td>—</td><td>—</td></tr><tr><td>OpenRouter ids</td><td>✅</td><td>—</td><td>—</td></tr><tr><td>Multi-model in one run</td><td>✅</td><td>—</td><td>—</td></tr><tr><td>Followup / lineage</td><td>✅ (OpenAI/Azure)</td><td>partial (multi-turn)</td><td>—</td></tr><tr><td>Image generation</td><td>—</td><td>✅</td><td>✅</td></tr><tr><td>YouTube analysis</td><td>—</td><td>—</td><td>✅</td></tr><tr><td>Deep Research</td><td>—</td><td>✅</td><td>—</td></tr><tr><td><code>--render --copy</code> fallback</td><td>✅</td><td>✅</td><td>✅</td></tr></tbody></table>
+<table><thead><tr><th>Capability</th><th>API</th><th>Browser (ChatGPT)</th><th>Browser (Gemini)</th></tr></thead><tbody><tr><td>GPT-5.x family</td><td>✅</td><td>✅</td><td>—</td></tr><tr><td>Gemini 3 / 3.1 Pro</td><td>✅</td><td>—</td><td>✅</td></tr><tr><td>Claude Sonnet / Opus</td><td>✅</td><td>—</td><td>—</td></tr><tr><td>OpenRouter ids</td><td>✅</td><td>—</td><td>—</td></tr><tr><td>Multi-model in one run</td><td>✅</td><td>—</td><td>—</td></tr><tr><td>Followup / lineage</td><td>✅ (OpenAI/Azure)</td><td>partial (multi-turn)</td><td>—</td></tr><tr><td>Image generation</td><td>—</td><td>✅</td><td>✅</td></tr><tr><td>YouTube analysis</td><td>—</td><td>—</td><td>✅</td></tr><tr><td>Deep Research</td><td>—</td><td>✅</td><td>—</td></tr><tr><td><code>--render --copy</code> fallback</td><td>✅</td><td>✅</td><td>✅</td></tr></tbody></table>
 <p>See provider-specific docs for the gory details: <a href="openai-endpoints.html">OpenAI / Azure / OpenRouter</a>, <a href="gemini.html">Gemini</a>, <a href="anthropic.html">Anthropic</a>, <a href="grok.html">Grok</a>.</p><nav class="page-nav" aria-label="Pager"><a class="page-nav-prev" href="configuration.html"><small>Previous</small><span>Local configuration (JSON5)</span></a><a class="page-nav-next" href="openai-endpoints.html"><small>Next</small><span>OpenAI-Compatible Endpoints</span></a></nav></article>
-        <nav class="toc" aria-label="On this page"><h2>On this page</h2><a class="toc-l2" href="#the-lineup">The lineup</a><a class="toc-l2" href="#when-to-reach-for-which">When to reach for which</a><a class="toc-l3" href="#gpt-5-5-pro-default">GPT-5.5 Pro (default)</a><a class="toc-l3" href="#gemini-3-1-pro">Gemini 3.1 Pro</a><a class="toc-l3" href="#claude-opus-4-1">Claude Opus 4.1</a><a class="toc-l3" href="#multi-model-triangulation">Multi-model triangulation</a><a class="toc-l2" href="#deep-research-browser-only">Deep Research (browser only)</a><a class="toc-l2" href="#thinking-time">Thinking time</a><a class="toc-l2" href="#pricing-notes-rough-heuristics-not-contracts">Pricing notes (rough heuristics, not contracts)</a><a class="toc-l2" href="#engine-compatibility">Engine compatibility</a></nav>
+        <nav class="toc" aria-label="On this page"><h2>On this page</h2><a class="toc-l2" href="#the-lineup">The lineup</a><a class="toc-l2" href="#when-to-reach-for-which">When to reach for which</a><a class="toc-l3" href="#gpt-5-5-pro-default">GPT-5.5 Pro (default)</a><a class="toc-l3" href="#gemini-3-pro">Gemini 3 Pro</a><a class="toc-l3" href="#claude-opus-4-1">Claude Opus 4.1</a><a class="toc-l3" href="#multi-model-triangulation">Multi-model triangulation</a><a class="toc-l2" href="#deep-research-browser-only">Deep Research (browser only)</a><a class="toc-l2" href="#thinking-time">Thinking time</a><a class="toc-l2" href="#pricing-notes-rough-heuristics-not-contracts">Pricing notes (rough heuristics, not contracts)</a><a class="toc-l2" href="#engine-compatibility">Engine compatibility</a></nav>
       </div>
     </main>
   </div>

package/dist/docs-site/openai-endpoints.html

@@ -285,17 +285,6 @@ body:not(.home) .doc>h1:first-child{display:none}
 <p>Force first-party OpenAI when Azure env vars are exported:</p>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --provider</span><span style="color:#96D0FF"> openai</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> api</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gpt-5.5-pro</span><span style="color:#6CB6FF"> -p</span><span style="color:#96D0FF"> "Review this"</span></span>
 <span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --no-azure</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> api</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gpt-5.5-pro</span><span style="color:#6CB6FF"> -p</span><span style="color:#96D0FF"> "Review this"</span></span></code></pre>
-<h2 id="gpt-5-6-pro-reasoning-mode"><a class="anchor" href="#gpt-5-6-pro-reasoning-mode" aria-label="Anchor link">#</a>GPT-5.6 Pro reasoning mode</h2>
-<p>GPT-5.6 Pro is a Responses API execution mode, not a separate model slug. Select the Sol model and enable Pro mode explicitly:</p>
-<pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> api</span><span style="color:#F47067"> \</span></span>
-<span class="line"><span style="color:#6CB6FF">  --model</span><span style="color:#96D0FF"> gpt-5.6-sol</span><span style="color:#F47067"> \</span></span>
-<span class="line"><span style="color:#6CB6FF">  --reasoning-mode</span><span style="color:#96D0FF"> pro</span><span style="color:#F47067"> \</span></span>
-<span class="line"><span style="color:#6CB6FF">  --reasoning-effort</span><span style="color:#96D0FF"> max</span><span style="color:#F47067"> \</span></span>
-<span class="line"><span style="color:#6CB6FF">  -p</span><span style="color:#96D0FF"> "Review this difficult architecture"</span><span style="color:#F47067"> \</span></span>
-<span class="line"><span style="color:#6CB6FF">  --file</span><span style="color:#96D0FF"> "src/**"</span></span></code></pre>
-<p>Oracle sends <code>model: &quot;gpt-5.6-sol&quot;</code> with <code>reasoning.mode: &quot;pro&quot;</code> and <code>reasoning.effort: &quot;max&quot;</code>. Mode and effort are independent. GPT-5.6 supports <code>none</code>, <code>low</code>, <code>medium</code>, <code>high</code>, <code>xhigh</code>, and <code>max</code>; without an explicit effort, Oracle retains the model&#39;s configured effort. Pro-mode runs use Oracle&#39;s long-running API defaults: they detach unless <code>--wait</code> is supplied, default to a 60-minute timeout, and use Responses background mode when the selected route supports it.</p>
-<p><code>--reasoning-mode</code> accepts <code>standard</code> or <code>pro</code>; <code>--reasoning-effort</code> accepts all six GPT-5.6 effort levels. Both are API-only and restricted to the GPT-5.6 family. Oracle rejects reasoning mode for OpenRouter and custom <code>--base-url</code> routes because those routes currently use Oracle&#39;s Chat Completions adapter, which cannot represent Responses API reasoning mode.</p>
-<p>Do not pass <code>--model gpt-5.6-pro</code> or <code>--model gpt-5.6-sol-pro</code>; Oracle rejects those fake slugs and points to <code>--reasoning-mode pro</code>.</p>
 <h2 id="custom-base-urls-litellm-localhost"><a class="anchor" href="#custom-base-urls-litellm-localhost" aria-label="Anchor link">#</a>Custom Base URLs (LiteLLM, Localhost)</h2>
 <p>For other compatible services that use the standard OpenAI protocol but a different URL:</p>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --base-url</span><span style="color:#96D0FF"> http://localhost:4000</span></span></code></pre>
@@ -303,18 +292,6 @@ body:not(.home) .doc>h1:first-child{display:none}
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-json"><span class="line"><span style="color:#ADBAC7">{</span></span>
 <span class="line"><span style="color:#8DDB8C">  "apiBaseUrl"</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"http://localhost:4000"</span></span>
 <span class="line"><span style="color:#ADBAC7">}</span></span></code></pre>
-<p>If a gateway exposes a different on-wire model id, add an API-only override to the user config at <code>~/.oracle/config.json</code>:</p>
-<pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-json5"><span class="line"><span style="color:#ADBAC7">{</span></span>
-<span class="line"><span style="color:#96D0FF">  apiBaseUrl</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"http://localhost:4000"</span><span style="color:#ADBAC7">,</span></span>
-<span class="line"><span style="color:#96D0FF">  model</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"gpt-5.5"</span><span style="color:#ADBAC7">,</span></span>
-<span class="line"><span style="color:#96D0FF">  modelOverrides</span><span style="color:#ADBAC7">: {</span></span>
-<span class="line"><span style="color:#96D0FF">    "gpt-5.5"</span><span style="color:#ADBAC7">: {</span></span>
-<span class="line"><span style="color:#96D0FF">      apiModel</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"gateway-model"</span><span style="color:#ADBAC7">,</span></span>
-<span class="line"><span style="color:#96D0FF">      reasoning</span><span style="color:#ADBAC7">: { </span><span style="color:#96D0FF">effort</span><span style="color:#ADBAC7">: </span><span style="color:#96D0FF">"high"</span><span style="color:#ADBAC7"> },</span></span>
-<span class="line"><span style="color:#ADBAC7">    },</span></span>
-<span class="line"><span style="color:#ADBAC7">  },</span></span>
-<span class="line"><span style="color:#ADBAC7">}</span></span></code></pre>
-<p>Overrides accept existing built-in model keys only and are ignored in project <code>.oracle/config.json</code> files. See <a href="configuration.html">Local configuration</a> for the full field list and precedence.</p>
 <h2 id="model-aliases"><a class="anchor" href="#model-aliases" aria-label="Anchor link">#</a>Model aliases</h2>
 <p>Oracle keeps a stable CLI-facing model set, but some names are aliases for the concrete API model ids it sends:</p>
 <ul>
@@ -344,7 +321,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <li>You can still set <code>--base-url</code> explicitly; if it points at OpenRouter (with or without a trailing <code>/responses</code>), Oracle will use <code>OPENROUTER_API_KEY</code> and forward optional attribution headers (<code>OPENROUTER_REFERER</code> / <code>OPENROUTER_TITLE</code>).</li>
 <li>Multi-model runs accept OpenRouter ids alongside built-in ones. See <code>docs/openrouter.md</code> for details.</li>
 </ul><nav class="page-nav" aria-label="Pager"><a class="page-nav-prev" href="mythical-pro-agents.html"><small>Previous</small><span>Mythical Pro Agents</span></a><a class="page-nav-next" href="anthropic.html"><small>Next</small><span>Anthropic (Claude) Integration Plan</span></a></nav></article>
-        <nav class="toc" aria-label="On this page"><h2>On this page</h2><a class="toc-l2" href="#azure-openai">Azure OpenAI</a><a class="toc-l2" href="#provider-diagnostics">Provider diagnostics</a><a class="toc-l3" href="#cli-configuration">CLI Configuration</a><a class="toc-l2" href="#gpt-5-6-pro-reasoning-mode">GPT-5.6 Pro reasoning mode</a><a class="toc-l2" href="#custom-base-urls-litellm-localhost">Custom Base URLs (LiteLLM, Localhost)</a><a class="toc-l2" href="#model-aliases">Model aliases</a><a class="toc-l3" href="#browser-engine-vs-api-base-urls">Browser engine vs API base URLs</a><a class="toc-l3" href="#example-litellm">Example: LiteLLM</a><a class="toc-l2" href="#openrouter">OpenRouter</a></nav>
+        <nav class="toc" aria-label="On this page"><h2>On this page</h2><a class="toc-l2" href="#azure-openai">Azure OpenAI</a><a class="toc-l2" href="#provider-diagnostics">Provider diagnostics</a><a class="toc-l3" href="#cli-configuration">CLI Configuration</a><a class="toc-l2" href="#custom-base-urls-litellm-localhost">Custom Base URLs (LiteLLM, Localhost)</a><a class="toc-l2" href="#model-aliases">Model aliases</a><a class="toc-l3" href="#browser-engine-vs-api-base-urls">Browser engine vs API base URLs</a><a class="toc-l3" href="#example-litellm">Example: LiteLLM</a><a class="toc-l2" href="#openrouter">OpenRouter</a></nav>
       </div>
     </main>
   </div>

package/dist/docs-site/quickstart.html

@@ -270,7 +270,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <p><code>--browser-manual-login</code> skips Keychain cookie copy (no permission popups) and reuses a persistent automation profile under <code>~/.oracle/browser/</code>.</p>
 <h3 id="render-and-copy"><a class="anchor" href="#render-and-copy" aria-label="Anchor link">#</a>Render and copy</h3>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --render</span><span style="color:#6CB6FF"> --copy</span><span style="color:#6CB6FF"> -p</span><span style="color:#96D0FF"> "Architecture review"</span><span style="color:#6CB6FF"> --file</span><span style="color:#96D0FF"> "src/**/*.ts"</span></span></code></pre>
-<p>The bundle is on your clipboard. Paste it into ChatGPT, Claude, Gemini, AI Studio, or wherever you want the answer. Generated text context includes stable <code>Lines:</code> ranges and <code>N |</code> prefixes for <code>path:line</code> citations. Direct browser file uploads and ZIP bundles keep the original file contents.</p>
+<p>The bundle is on your clipboard. Paste it into ChatGPT, Claude, Gemini, AI Studio, or wherever you want the answer.</p>
 <h2 id="3-preview-before-you-spend"><a class="anchor" href="#3-preview-before-you-spend" aria-label="Anchor link">#</a>3. Preview before you spend</h2>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --dry-run</span><span style="color:#96D0FF"> summary</span><span style="color:#6CB6FF"> --files-report</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  -p</span><span style="color:#96D0FF"> "Audit the storage layer for race conditions"</span><span style="color:#F47067"> \</span></span>

package/dist/docs-site/sessions.html

@@ -251,7 +251,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <span class="line"><span>├── prompt.md                # assembled bundle (what was sent)</span></span>
 <span class="line"><span>├── response.md              # the model's answer (when complete)</span></span>
 <span class="line"><span>├── log.jsonl                # per-event log</span></span>
-<span class="line"><span>└── artifacts/               # browser-only: transcript, generated images/files, deep-research-report.md</span></span></code></pre>
+<span class="line"><span>└── artifacts/               # browser-only: transcript, generated images, deep-research-report.md</span></span></code></pre>
 <p>Override the root with <code>ORACLE_HOME_DIR=/some/path</code>.</p>
 <h2 id="listing"><a class="anchor" href="#listing" aria-label="Anchor link">#</a>Listing</h2>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#96D0FF"> status</span><span style="color:#768390">                  # last 20 sessions</span></span>
@@ -277,10 +277,9 @@ body:not(.home) .doc>h1:first-child{display:none}
 <span class="line"><span>Models: 3 parallel</span></span>
 <span class="line"><span>Detach: yes, polling</span></span>
 <span class="line"><span>Reattach: oracle session 20260515-name-panel</span></span></code></pre>
-<p><code>oracle status</code> uses compact mode labels such as <code>api/fg</code>, <code>api/bg</code>, <code>br/fg</code>, and <code>br/bg</code>; <code>oracle session &lt;id&gt;</code> shows the persisted execution state.</p>
-<p>To keep the original CLI attached until completion, pass <code>--wait</code>:</p>
+<p><code>oracle status</code> uses compact mode labels such as <code>api/fg</code>, <code>api/bg</code>, and <code>br/fg</code>; <code>oracle session &lt;id&gt;</code> shows the persisted execution state.</p>
+<p>To block in the original command, pass <code>--wait</code>:</p>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --wait</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gpt-5.5-pro</span><span style="color:#6CB6FF"> -p</span><span style="color:#96D0FF"> "Long architecture review"</span><span style="color:#6CB6FF"> --file</span><span style="color:#96D0FF"> "src/**"</span></span></code></pre>
-<p>For API runs, <code>--wait</code> executes the request in the foreground. Local Pro browser runs use a detached worker even with <code>--wait</code>, while the original CLI stays attached to the session log. This lets the browser worker capture and save the answer if the foreground CLI exits unexpectedly. Pressing Ctrl-C still cancels the worker and exits with code 130.</p>
 <p>For browser runs, ChatGPT sometimes redirects mid-page-load. The auto-reattach flags poll the existing tab without manual intervention:</p>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --engine</span><span style="color:#96D0FF"> browser</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --browser-timeout</span><span style="color:#96D0FF"> 6m</span><span style="color:#F47067"> \</span></span>
@@ -293,10 +292,10 @@ body:not(.home) .doc>h1:first-child{display:none}
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#96D0FF"> restart</span><span style="color:#F47067"> &#x3C;</span><span style="color:#96D0FF">i</span><span style="color:#ADBAC7">d</span><span style="color:#F47067">></span><span style="color:#768390">            # re-run with the same prompt + files</span></span></code></pre>
 <p>Useful when a transient browser/API error truncated the answer. Restart copies the bundle, opens a new session, and links lineage back.</p>
 <h2 id="follow-up"><a class="anchor" href="#follow-up" aria-label="Anchor link">#</a>Follow up</h2>
-<p>Continue a saved ChatGPT browser conversation or an OpenAI / Azure Responses API session with new context:</p>
+<p>Continue an OpenAI / Azure Responses API session with new context:</p>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --followup</span><span style="color:#F47067"> &#x3C;</span><span style="color:#96D0FF">i</span><span style="color:#ADBAC7">d</span><span style="color:#F47067">></span><span style="color:#6CB6FF"> -p</span><span style="color:#96D0FF"> "Re-evaluate with these files"</span><span style="color:#F47067"> \</span></span>
 <span class="line"><span style="color:#6CB6FF">  --file</span><span style="color:#96D0FF"> "src/migrations/**"</span></span></code></pre>
-<p>Browser followup reopens the exact saved conversation and inherits its browser configuration and model. For multi-model API parents, pick the lineage with <code>--followup-model</code>. See <a href="followup.html">Followup</a> for the full flow and the formats <code>--followup</code> accepts (session ids, slugs, or <code>resp_…</code> response ids).</p>
+<p>For multi-model parents, pick the lineage with <code>--followup-model</code>. See <a href="followup.html">Followup</a> for the full flow and the formats <code>--followup</code> accepts (session ids, slugs, or <code>resp_…</code> response ids).</p>
 <h2 id="background-mode"><a class="anchor" href="#background-mode" aria-label="Anchor link">#</a>Background mode</h2>
 <p>Force a Responses API run into background mode (create + retrieve) regardless of model defaults:</p>
 <pre class="shiki github-dark-dimmed" style="background-color:var(--code-bg);color:var(--code-fg)" tabindex="0"><code class="language-bash"><span class="line"><span style="color:#F69D50">oracle</span><span style="color:#6CB6FF"> --background</span><span style="color:#6CB6FF"> --model</span><span style="color:#96D0FF"> gpt-5.5-pro</span><span style="color:#6CB6FF"> -p</span><span style="color:#96D0FF"> "..."</span><span style="color:#6CB6FF"> --file</span><span style="color:#96D0FF"> "src/**"</span></span>

package/dist/docs-site/spec.html

@@ -275,7 +275,7 @@ body:not(.home) .doc>h1:first-child{display:none}
 <p>Auto-pick rules:</p>
 <ol>
 <li>If <code>--engine</code> is set, use it.</li>
-<li>Else if <code>engine</code> is set in the effective config, use it. The effective config starts with <code>~/.oracle/config.json</code>, then layers project <code>.oracle/config.json</code> files from parent folders to the current directory.</li>
+<li>Else if <code>engine</code> is set in <code>~/.oracle/config.json</code>, use it.</li>
 <li>Else if <code>OPENAI_API_KEY</code> (or another supported API key) is set, use API.</li>
 <li>Else use browser.</li>
 </ol>

package/dist/docs-site/windows-work.html

@@ -252,7 +252,6 @@ body:not(.home) .doc>h1:first-child{display:none}
 <li>agent-scripts bash helpers: <code>runner</code>/<code>scripts/committer</code> can fail under PowerShell/CMD because of CRLF and bash expectations. If they explode, run commands directly (<code>pnpm ...</code>, <code>git add/commit</code>) instead.</li>
 <li>browser-tools binary: not built in <code>agent-scripts/bin</code> on Windows; <code>pnpm tsx scripts/browser-tools.ts</code> also fails there (no package manifest). Use a macOS-built binary or run from macOS if you need it.</li>
 <li>Prefer PowerShell + pnpm directly; watch for CRLF warnings when touching tracked files.</li>
-<li>WSL browser launch host detection: a systemd-resolved stub such as <code>nameserver 127.0.0.53</code> is guest loopback, not the Windows host. Keep resolver-derived non-loopback hosts for Windows Chrome compatibility, but route resolver-derived <code>127/8</code> values to the standard local Chrome launcher.</li>
 </ul>
 <p>Future Windows gotchas belong here. Update this doc when you learn something new.</p>
 <ul>

package/dist/markdansi/types/index.js

@@ -0,0 +1,4 @@
+// Public API typings for Markdansi.
+// Hand-authored source for `pnpm types` (tsc --emitDeclarationOnly)
+// to produce dist/index.d.ts. Keep in sync with src/ changes.
+export {};