@stage5/lumine 0.3.51 → 0.3.53

package/lib/build-review.js

@@ -761,8 +761,21 @@ async function runChromeReview({
   });
   const closed = new Promise((resolve, reject) => {
     child.once("error", reject);
-    child.once("close", (code, signal) => {
-      resolve({ code: Number(code ?? -1), signal: signal || null });
+    child.once("exit", (code, signal) => {
+      // Chrome's crash reporter can inherit stderr and outlive Chrome. Node's
+      // `close` then never fires even though the browser already exited.
+      // Drain the last output briefly, then close only our pipe endpoints;
+      // do not kill another process or turn an early browser exit into success.
+      let done = false;
+      const finish = (stdioDrainTimedOut) => {
+        if (done) return;
+        done = true;
+        clearTimeout(drainTimer);
+        for (const stream of child.stdio) stream?.destroy?.();
+        resolve({ code: Number(code ?? -1), signal: signal || null, stdioDrainTimedOut });
+      };
+      const drainTimer = setTimeout(() => finish(true), 250);
+      child.once("close", () => finish(false));
     });
   });
   // The hard kill covers the observation window plus, when a script runs,
@@ -1140,6 +1153,7 @@ export async function runManagedBuildReview({ options, authToken, buildId }) {
       closeRequested: browser.closeRequested === true,
       forcedTermination: browser.forcedTermination === true,
       exitedCleanly: browser.code === 0 && !browser.signal,
+      stdioDrainTimedOut: browser.stdioDrainTimedOut === true,
       evidenceCompleteBeforeClose: browser.evidenceCompleteBeforeClose === true,
       exitedBeforeClose: browser.exitedBeforeClose === true,
       closeDurationMs: Number.isFinite(browser.closeDurationMs)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stage5/lumine",
-  "version": "0.3.51",
+  "version": "0.3.53",
   "description": "Command line tools for launching Lumine builds on Twinkle.",
   "type": "module",
   "bin": {

package/sdk/BUILD_SDK_INDEX.md

@@ -2,7 +2,7 @@
 
 Version: 1.68.0
 Updated: 2026-10-16
-Generated: 2026-10-19T05:26:13.264Z
+Generated: 2026-10-21T05:16:33.397Z
 
 ## Notes
 - This SDK is injected into Build iframes via the Build preview/runtime.
@@ -508,9 +508,9 @@ const result = await Twinkle.ai.chat({ message, history: chatHistory, systemProm
   - Template values describe types, not answers to copy. All object keys are required and extra keys are rejected, including inside nested objects; {} means an exactly empty object, not a free-form dictionary. A non-empty array uses only its first item as the template for every returned item, with no fixed length; mixed examples do not declare a union. [] allows arbitrary JSON items. A null template accepts any JSON primitive (string, number, boolean or null), but the key is still required. Use 'number' for fractional numbers; 0 and other integer examples require integers. State content requirements in prompt or instructions. GPT routes retain descriptive string examples as field guidance, not enforced enum values.
   - GPT models prefer native JSON Schema output for compatible shapes. Shapes outside the strict provider subset, such as [] with arbitrary items or deeply nested structures, use prompt-based JSON generation with the same final validation. A provider rejection specifically of the native schema triggers one prompt-based fallback; auth, quota, safety and unrelated request failures do not trigger that fallback. Validation retries preserve the original task and instructions. The returned object/result shape and streaming callbacks are unchanged.
   - mode is accepted as an alias for thinkingMode, and mid is accepted as an alias for medium.
-  - Omit model to use the normal Lite/Medium/High routing. Lumine Light defaults to claude-haiku-5-5; gpt-6-luna remains its second choice. Both model IDs accept Low, Medium or High thinking. Explicit claude-sonnet-5-5, gpt-6.1-sol or claude-opus-5-5 still require thinkingMode: 'high'; unknown model IDs reject instead of silently falling back. Retired IDs still work and run on their replacement: claude-sonnet-5 runs claude-sonnet-5-5; gpt-6-sol and gpt-5.6-sol run gpt-6.1-sol; gpt-6-astra, claude-opus-5 and claude-fable-5-1 run claude-opus-5-5.
-  - thinkingMode low uses Claude Haiku 5.5 with low adaptive thinking and consumes the viewer's AI Energy from confirmed provider usage. Set model: 'gpt-6-luna' to choose Luna instead.
-  - thinkingMode medium is Lumine Light: it uses Claude Haiku 5.5 with medium adaptive thinking and consumes normal AI Energy. Set model: 'gpt-6-luna' to keep Luna as the second choice.
+  - Omit model to use the normal Lite/Medium/High routing. Low and Medium use gpt-6-luna; explicit gpt-6-luna accepts Low, Medium or High thinking. Existing apps naming claude-haiku-5-5 now run on gpt-6-luna at the requested thinking level. Explicit claude-sonnet-5-5, gpt-6.1-sol or claude-opus-5-5 still require thinkingMode: 'high'; unknown model IDs reject instead of silently falling back. Retired IDs still work and run on their replacement: claude-sonnet-5 runs claude-sonnet-5-5; gpt-6-sol and gpt-5.6-sol run gpt-6.1-sol; gpt-6-astra, claude-opus-5 and claude-fable-5-1 run claude-opus-5-5.
+  - thinkingMode low uses GPT-6 Luna with low reasoning and consumes the viewer's AI Energy from confirmed provider usage.
+  - thinkingMode medium uses GPT-6 Luna with medium reasoning and consumes normal AI Energy.
   - thinkingMode high without model uses Claude Sonnet 5.5 with high adaptive thinking and consumes high AI Energy. Explicit model: 'gpt-6.1-sol' selects GPT-6.1 Sol with xhigh reasoning at the same High AI Energy tier; model: 'claude-sonnet-5-5' names Sonnet 5.5 directly.
   - claude-sonnet-5-5 and claude-opus-5-5 use Anthropic adaptive High thinking and debit confirmed provider usage at the High tier.
   - Pass onStatus, onReasoning, and/or onText to stream progress from the same structured generation. onStatus receives high-level phases such as thinking, searching_web, responding, validating, and completed.
@@ -519,7 +519,7 @@ const result = await Twinkle.ai.chat({ message, history: chatHistory, systemProm
   - Put a user-facing field such as producerNotes in expectedStructure when commentary must be part of the validated final object rather than transient reasoning progress.
   - When AI Energy is empty, every automatic or named model choice rejects before new provider work; there is no free fallback mode.
   - Live web search is enabled by default in Medium and High modes. Pass webSearch: false to disable it for the app. Low/Lite Mode remains tool-free; explicitly forcing webSearch: true in Low Mode returns an error.
-  - The server validates the final shape; OpenAI/xAI routes can retry malformed output, while Anthropic routes (claude-haiku-5-5, claude-sonnet-5-5 and claude-opus-5-5) use native JSON Schema output and retry one malformed or shape-invalid result. App code should still validate business-specific enum values.
+  - The server validates the final shape; OpenAI/xAI routes can retry malformed output, while Anthropic routes (claude-sonnet-5-5 and claude-opus-5-5) use native JSON Schema output and retry one malformed or shape-invalid result. App code should still validate business-specific enum values.
   - If the model declines a request, the call rejects with code ai_declined (HTTP 422). The tokens it used are charged to AI Energy like a finished call. Show the viewer a short message so they can rephrase; do not automatically retry the same prompt.
   - Example: const { object } = await Twinkle.ai.generateObject({ thinkingMode: 'high', model: 'claude-opus-5-5', prompt: 'Plan the next section from: ' + currentState, expectedStructure: { producerNotes: 'string', action: 'string', confidence: 'number' }, onStatus: (phase) => showPhase(phase), onReasoning: (summary, meta) => showReasoningProgress(summary, meta), onText: (partialJson, meta) => showStructuredProgress(partialJson, meta) });
 - onChatStatus(listener) | scopes: none
@@ -601,9 +601,9 @@ const result = await Twinkle.ai.chat({ message, history: chatHistory, systemProm
   - The character route also accepts text or message fields for compatibility, but generated apps should use content.
   - The server keeps the latest 16 valid character history entries.
   - Pass onText/onStatus for streaming dialogue. Omit callbacks for non-streaming dialogue where the promise resolves with the final response.
-  - Inside Build character chat, thinkingMode low uses Lite Mode: Zero uses GPT-6 Luna with reasoning disabled and Ciel uses Claude Haiku 5.5 with low thinking; confirmed provider usage consumes the viewer's AI Energy and is usually cheaper than High.
-  - Inside Build character chat, thinkingMode medium uses the same normal chat model routing: Zero uses GPT-6 Luna with reasoning disabled and Ciel uses Claude Haiku 5.5 with low thinking, with normal AI Energy. If Anthropic is unavailable, Ciel's reply falls back to GPT-6 Luna.
-  - Inside Build character chat, thinkingMode high uses Think Hard chat routing and high AI Energy: Zero uses Grok 4.7 with high reasoning and Ciel uses Claude Opus 5.5 with high thinking.
+  - Inside Build character chat, thinkingMode low uses Lite Mode: Zero uses GPT-6 Luna with reasoning disabled and Ciel uses GPT-6 Luna with reasoning disabled; confirmed provider usage consumes the viewer's AI Energy and is usually cheaper than High.
+  - Inside Build character chat, thinkingMode medium uses the same normal chat model routing: Zero uses GPT-6 Luna with reasoning disabled and Ciel uses GPT-6 Luna with reasoning disabled, with normal AI Energy.
+  - Inside Build character chat, thinkingMode high uses Think Hard chat routing and high AI Energy: Zero uses Grok 4.7 with high reasoning and Ciel uses Claude Opus 5.5 with high thinking, with GPT-6 Luna at high reasoning for recovery.
   - When AI Energy is empty, Low, Medium, and High all reject before new provider work; there is no free fallback mode.
   - Pass roomContext as a short shared scene transcript so Zero and Ciel can know what happened in the same room.
   - includeWebsiteContext defaults to true. Set includeWebsiteContext: false for in-world NPC dialogue that should only use Zero/Ciel's basic character identity plus your scene/instructions.

package/sdk/LUMINE_ADMIN.md

@@ -1184,7 +1184,7 @@ lumine admin report serve --file <dir>/daily-management-report.md [--port <n>] [
 ```
 
 These compose the documented commands in this contract, running them as child processes, so every scope check, audit and pagination rule applies unchanged.
-- **`gather`** saves each read-only report into `<dir>`, including the canonical published Math Lab queue in `math-lab-queue.json`. It follows `bot-output` to exhaustion and renders `bot-chats.txt`, then lists any failed step in `gather.json`. Total/per-step `durationMs`, attempts and recovered reads make recurring slow or failing collection steps visible during the daily review. It also creates `review-state.json`: every source is fetched, not read. Children bypass update checks, require matching fresh stdout/file receipts, and retry an unconfirmed opted-in read at most once. Mutations never auto-retry. It needs the active run for run-scoped reads.
+- **`gather`** saves its fixed set of read-only reports into `<dir>`, including the canonical published Math Lab queue in `math-lab-queue.json`. It is not the complete daily checklist: app refills/usage, all-type Build reviews, sponsor integrity and other duties need their own steps. Repository agents follow the current full-run coverage index in `agent-guides/lumine-admin.md`. It follows `bot-output` to exhaustion and renders `bot-chats.txt`, then lists any failed step in `gather.json`. Total/per-step `durationMs`, attempts and recovered reads make recurring slow or failing collection steps visible during the daily review. It also creates `review-state.json`: every source is fetched, not read. Children bypass update checks, require matching fresh stdout/file receipts, and retry an unconfirmed opted-in read at most once. Mutations never auto-retry. It needs the active run for run-scoped reads.
 - **`progress`** reads local review state and explicit operator acknowledgments, writes private `progress.json`, and keeps reading, canonical effort mutations, the run lease, and browser checks separate. Add all required duties for the actual run scope and its carryovers; collection alone is not a full review. A completed run lease cannot complete a pending duty. Reading acknowledgments bind to the current file bytes and are operator attestations, not proof of comprehension.
 - **`featured candidates`** returns Subjects posted after the Bangkok-midnight cutoff that are off the board and provably never Featured. Each one comes with its human/bot comment counts, explicit comment truncation/completeness, and whether Mikey has viewed it. Counts from truncated context are partial. An unavailable Subject context fails the collection. Eligibility only: ranking stays editorial.
 - **`report serve`** detaches a loopback server that answers only `/`, `marked.min.js` and the report Markdown. `--open` opens Chrome and reports whether the OS accepted the request. Render, navigation and original-source verification remain pending until actually checked. Startup failures stop only the child this command started.