mixdog 0.9.174 → 1.0.0

package/scripts/plan-runtime-release-tag.mjs

@@ -0,0 +1,57 @@
+// Decides the release tag a memory runtime build publishes under.
+//
+// Published runtime assets are immutable: every app that shipped a manifest
+// naming a tag verifies that tag's assets by size and sha256, so replacing an
+// asset under a shipped tag breaks the next runtime install of all of those
+// apps. A build that would replace an asset of a shipped tag (any tag at or
+// below the bundled manifest's release_tag) moves to the next patch tag; the
+// manifest sync then points new app builds at it.
+//
+// CLI: node scripts/plan-runtime-release-tag.mjs <requested-tag> <shipped-tag> <needed-assets-file> <published-assets-file>
+// prints the tag to build. The asset files hold one asset name per line.
+import { readFileSync } from 'node:fs';
+import { pathToFileURL } from 'node:url';
+
+const TAG_RE = /^runtime-v(\d+)\.(\d+)\.(\d+)$/;
+
+function parseTag(tag) {
+  const match = TAG_RE.exec(String(tag ?? '').trim());
+  if (!match) throw new Error(`invalid runtime release tag: ${tag}`);
+  return match.slice(1).map(Number);
+}
+
+function compareTags(left, right) {
+  const a = parseTag(left);
+  const b = parseTag(right);
+  for (let index = 0; index < 3; index += 1) {
+    if (a[index] !== b[index]) return a[index] - b[index];
+  }
+  return 0;
+}
+
+export function planRuntimeReleaseTag({ requestedTag, shippedTag, neededAssets, publishedAssets }) {
+  const published = new Set(publishedAssets);
+  const replacesPublished = neededAssets.some((name) => published.has(name));
+  if (!replacesPublished || compareTags(requestedTag, shippedTag) > 0) return requestedTag;
+  const [major, minor, patch] = parseTag(shippedTag);
+  return `runtime-v${major}.${minor}.${patch + 1}`;
+}
+
+function readNames(path) {
+  return readFileSync(path, 'utf8')
+    .split(/\r?\n/)
+    .map((line) => line.trim())
+    .filter(Boolean);
+}
+
+if (import.meta.url === pathToFileURL(process.argv[1] || '').href) {
+  const [requestedTag, shippedTag, neededPath, publishedPath] = process.argv.slice(2);
+  process.stdout.write(
+    `${planRuntimeReleaseTag({
+      requestedTag,
+      shippedTag,
+      neededAssets: readNames(neededPath),
+      publishedAssets: readNames(publishedPath),
+    })}\n`
+  );
+}

package/scripts/prepare-native-assets.mjs

@@ -56,7 +56,7 @@ export async function prepareRequiredNativeAssets({
       }
       const destination = join(stagedToolsDir, fileName);
       await copyFile(installedPath, destination);
-      if (process.platform !== 'win32') {
+      if (platform !== 'win32') {
         await chmod(destination, 0o755);
       }
       return [name, join(target, fileName)];

package/scripts/provider-contract-test-env.mjs

@@ -0,0 +1,21 @@
+// Import this FIRST in provider-contract-test.mjs. Runtime modules resolve the
+// data dir (and the config file under it) once, at load, so the isolated dir
+// must exist before any of them is imported; setting it later leaves the
+// developer's real config — including Settings → Developer options — in play.
+//
+// Usage snapshots persist to <data dir>/gateway-oauth-usage-cache.json. Without
+// an isolated data dir these fixtures wrote provider rows into the developer's
+// real cache, where a fake model id could later win a provider fallback lookup.
+import { mkdtempSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+
+const PROVIDER_CONTRACT_DATA_DIR = mkdtempSync(join(tmpdir(), 'mixdog-provider-contract-'));
+process.env.MIXDOG_DATA_DIR = PROVIDER_CONTRACT_DATA_DIR;
+process.on('exit', () => {
+  try {
+    rmSync(PROVIDER_CONTRACT_DATA_DIR, { recursive: true, force: true, maxRetries: 5, retryDelay: 50 });
+  } catch {
+    /* a still-open handle leaves it for the OS temp cleanup */
+  }
+});

package/scripts/release-paths.mjs

@@ -22,6 +22,7 @@ export const DESKTOP_GATE_PREFIXES = ['apps/desktop/', 'src/', 'vendor/', 'LICEN
 export const DESKTOP_GATE_FILES = ['README.md', 'NOTICE.md', ...PACKAGE_MANIFESTS];
 // Desktop packaging/postinstall scripts; `.test` companions ride along.
 export const DESKTOP_GATE_SCRIPT_BASES = [
+  'scripts/prune-desktop-runtime',
   'scripts/prune-embedding-runtime',
   'scripts/native-binary-arch',
   'scripts/native-tool-download',

package/scripts/routing-corpus.mjs

@@ -9,6 +9,7 @@ import { basename, resolve, win32 } from 'node:path';
 import { resolvePluginData } from '../src/runtime/shared/plugin-paths.mjs';
 import { parseSince } from './lib/parse-since.mjs';
 import { argValue, hasFlag } from './lib/cli-args.mjs';
+import { field, sessionId } from './lib/trace-row.mjs';
 
 function defaultTracePath() {
   const data = process.env.MIXDOG_DATA_DIR || resolvePluginData() || resolve(homedir(), '.mixdog', 'data');
@@ -28,17 +29,6 @@ function readRows(path) {
   }
   return out;
 }
-function payload(row) {
-  return row?.payload && typeof row.payload === 'object' ? row.payload : {};
-}
-function field(row, name) {
-  if (row && row[name] != null) return row[name];
-  const p = payload(row);
-  return p[name] != null ? p[name] : null;
-}
-function sessionId(row) {
-  return String(row?.session_id || row?.sessionId || field(row, 'session_id') || '');
-}
 function shortId(id) {
   const s = String(id || '');
   return s.length <= 18 ? s : `${s.slice(0, 10)}…${s.slice(-6)}`;

package/scripts/session-diag.mjs

@@ -1,10 +1,9 @@
 #!/usr/bin/env node
 import { existsSync, readFileSync } from 'node:fs';
-import { homedir } from 'node:os';
-import { resolve } from 'node:path';
 import { parseSince } from './lib/parse-since.mjs';
 import { argValue, intArg } from './lib/cli-args.mjs';
 import { stats } from './lib/trace-stats.mjs';
+import { defaultTraceFiles, field, num as numberField } from './lib/trace-row.mjs';
 import { isInclusiveProvider } from '../src/runtime/shared/llm/cost.mjs';
 
 const pathArg = argValue('--path', null);
@@ -16,32 +15,6 @@ const limit = intArg('--limit', 30);
 const jsonMode = process.argv.includes('--json');
 const treeMode = process.argv.includes('--tree');
 
-const mixdogHome = process.env.MIXDOG_HOME || resolve(homedir(), '.mixdog');
-const mixdogDataDir = process.env.MIXDOG_DATA_DIR || resolve(mixdogHome, 'data');
-
-function unique(values) {
-  const seen = new Set();
-  const out = [];
-  for (const value of values) {
-    const key = String(value || '');
-    if (!key || seen.has(key)) continue;
-    seen.add(key);
-    out.push(value);
-  }
-  return out;
-}
-
-function defaultTraceFiles() {
-  if (pathArg) return [resolve(pathArg)];
-  const dirs = dataDir ? [resolve(dataDir)] : [resolve(process.cwd(), '.mixdog', 'data'), mixdogDataDir];
-  return unique(
-    dirs.flatMap((dir) => [
-      resolve(dir, 'history', 'agent-trace.jsonl.1'),
-      resolve(dir, 'history', 'agent-trace.jsonl'),
-    ])
-  );
-}
-
 function readRows(file) {
   if (!existsSync(file)) return [];
   return readFileSync(file, 'utf8')
@@ -56,21 +29,6 @@ function readRows(file) {
     });
 }
 
-function payload(row) {
-  return row?.payload && typeof row.payload === 'object' ? row.payload : {};
-}
-
-function field(row, name) {
-  if (row && row[name] != null) return row[name];
-  const p = payload(row);
-  return p[name] != null ? p[name] : null;
-}
-
-function numberField(row, name) {
-  const n = Number(field(row, name));
-  return Number.isFinite(n) ? n : null;
-}
-
 function values(nums) {
   return nums.filter((n) => Number.isFinite(n));
 }
@@ -386,7 +344,7 @@ function findSessionsByQuery(summaries, query) {
   return summaries.filter((s) => sessionMatchesQuery(s.id, query));
 }
 
-const files = defaultTraceFiles();
+const files = defaultTraceFiles({ pathArg, dataDir });
 const sinceTs = parseSince(sinceArg);
 const allRows = files
   .flatMap(readRows)

package/scripts/smoke-runtime-negative.sh

@@ -42,9 +42,15 @@ curl -sIL -o /dev/null -w "  HTTP %{http_code}\n" "$URL" || {
 }
 
 echo "==> Test 2: full download + sha256 (manifest match)"
-curl -sL -o "$WORK/$ASSET" "$URL"
+curl -fsSL -o "$WORK/$ASSET" "$URL"
 SHA="$(sha256sum "$WORK/$ASSET" 2>/dev/null | awk '{print $1}' || shasum -a 256 "$WORK/$ASSET" | awk '{print $1}')"
 echo "  sha256=$SHA"
+EXPECTED_SHA="$(curl -fsSL "${URL}.sha256" | awk '{print tolower($1)}')"
+[[ "$SHA" == "$EXPECTED_SHA" ]] || {
+	echo "FAIL: sha256 mismatch (got $SHA, release sidecar says ${EXPECTED_SHA:-<empty>})"
+	exit 1
+}
+echo "  PASS: sha256 matches release sidecar"
 
 echo "==> Test 3: corrupt tarball — flip one byte and ensure tar fails or runtime detects"
 cp "$WORK/$ASSET" "$WORK/corrupt.tar.gz"

package/scripts/test-data-dir.mjs

@@ -0,0 +1,19 @@
+// Side-effect module: import FIRST in tests that touch the read-snapshot store.
+// Points the Mixdog data dir at a per-run temp dir before snapshot-store.mjs
+// resolves SNAPSHOT_DIR at module load, so tests never write to the real one.
+import { mkdtempSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+
+if (!process.env.MIXDOG_TEST_DATA_DIR_OWNED) {
+  const dir = mkdtempSync(join(tmpdir(), 'mixdog-test-data-'));
+  process.env.MIXDOG_DATA_DIR = dir;
+  process.env.MIXDOG_TEST_DATA_DIR_OWNED = '1';
+  process.on('exit', () => {
+    try {
+      rmSync(dir, { recursive: true, force: true });
+    } catch {
+      /* best effort */
+    }
+  });
+}

package/scripts/tidy/sync-engines-manifest.mjs

@@ -334,7 +334,7 @@ function clangFormatAssets(release) {
   };
   const assets = {};
   for (const [platform, name] of Object.entries(map)) {
-    assets[platform] = { match: (candidate) => candidate === name };
+    assets[platform] = exact(name);
   }
   return { version: `${major}.0.0`, assets };
 }
@@ -412,8 +412,8 @@ function listTarGz(buffer) {
         const real = tar.subarray(offset, offset + 512);
         const realSize = Number.parseInt(real.subarray(124, 136).toString('utf8').replace(/\0.*$/, '').trim(), 8) || 0;
         offset += 512 + Math.ceil(realSize / 512) * 512;
-        continue;
       }
+      continue;
     } else if (name) {
       names.push(name);
     }
@@ -430,10 +430,7 @@ function pickBinPath(entries, binNames, fallback) {
   }
   const loose = files.filter((file) => {
     const base = file.split('/').pop();
-    return binNames.some((want) => {
-      const stem = want.replace(/\.exe$/i, '');
-      return base === want || base.startsWith(`${stem}-`) || base.startsWith(`${stem}_`) || base.startsWith(stem);
-    });
+    return binNames.some((want) => base.startsWith(want.replace(/\.exe$/i, '')));
   });
   if (loose.length === 1) return loose[0];
   if (fallback && files.includes(fallback)) return fallback;
@@ -706,11 +703,12 @@ async function main() {
       throw new Error(`--pin unknown engine '${id}'`);
     }
   }
-  if (checkOnly) {
-    await check();
-    return;
+  try {
+    if (checkOnly) await check();
+    else await generate(pins);
+  } finally {
+    await rm(TMP_DIR, { recursive: true, force: true });
   }
-  await generate(pins);
 }
 
 main().catch((error) => {

package/scripts/tool-efficiency-diag.mjs

@@ -4,6 +4,7 @@
 import { readFileSync, existsSync } from 'node:fs';
 import { homedir } from 'node:os';
 import { resolve } from 'node:path';
+import { field as F } from './lib/trace-row.mjs';
 
 function sinceArg() {
   const i = process.argv.indexOf('--since');
@@ -32,10 +33,6 @@ for (const f of files) {
     } catch {}
   }
 }
-const F = (r, n) => {
-  if (r[n] != null) return r[n];
-  return r.payload && r.payload[n] != null ? r.payload[n] : null;
-};
 const bySess = new Map();
 for (const r of rows) {
   const k = r.sessionId || r.session_id || '?';

package/scripts/voice-runtime-config.json

@@ -1,5 +1,5 @@
 {
-  "runtimeVersion": "1.10.0",
+  "runtimeVersion": "1.10.1",
   "whisperCommit": "306c88f4d1286aec1bf96e544632897886af5501",
   "ffmpegVersion": "6.1.1-mixdog.1",
   "vulkanSdkVersion": "1.4.309.0",

package/src/app.mjs

@@ -29,7 +29,7 @@ function warnWhenTuiBundleStale(bundle) {
     const hotSources = [
       join(__dirname, 'runtime', 'agent', 'orchestrator', 'session', 'agent-loop.mjs'),
       join(__dirname, 'runtime', 'agent', 'orchestrator', 'session', 'loop', 'stored-tool-args.mjs'),
-      join(__dirname, 'standalone', 'agent-tool.mjs'),
+      join(__dirname, 'session-runtime', 'services', 'agent-tool.mjs'),
     ];
     // The TUI half of the guard used to sample ONE path that does not exist
     // (tui/engine/session-api-ext.mjs), so every real src/tui edit ran silently

package/src/defaults/skills/code-tidy/SKILL.md

@@ -181,7 +181,9 @@ next step needs.
      reported, never auto-applied. Within a tier: comments → dead code →
      defensive code → duplication → complexity → abstraction → performance.
      Use targeted checks at meaningful behavior boundaries and run the
-     documented final tests/typecheck once after the round. Do not rerun
+     documented final tests/typecheck once after the round. Targeted checks
+     include the tests that import each edited module (found with `code_graph`
+     dependents or grep). Do not rerun
      unaffected checks after every comment or mechanical edit.
    - Structural items from the table below are CAREFUL unless their row says
      otherwise; dead code follows
@@ -264,7 +266,8 @@ An unperformed confirmed split stays unfinished.
 ## 7. Round close and final report
 Close each round in the conversation with a few lines: the partition, its test
 lane, what landed, the verification result including every check that could not
-run, the round's function counts over 50, 100 and 150 lines, and whether the
+run, the round's function counts over 50, 100 and 150 lines (taken from the
+function-length summary in the `tidy` check/fix result), and whether the
 round is complete or unfinished. Report all three counts: decomposition moves
 mass downward, so functions over 100 falling while functions over 50 rise is
 progress, and a single threshold hides it. Keep the inventory in `references/cleanup-report.md` form as you go;

package/src/defaults/skills/pptx/SKILL.md

@@ -105,7 +105,7 @@ The runtime absorbs what it can (one `a:pPr` per paragraph; package validation a
 | `charSpacing`, not `letterSpacing` | ignored silently |
 | Bullets: `bullet: true` on each item's first run, `breakLine: true` on every item but the last; `paraSpaceAfter`, not `lineSpacing`, between items | literal bullet characters and lineSpacing double the spacing |
 | `rectRadius` only on `roundRect`; other corners are their own presets | ignored silently |
-| Solid shape fills only; a gradient is a rasterized SVG image (`gradientField`) | gradient fill options are dropped |
+| Solid shape fills only; a gradient is drawn with the kit's `gradient()` (saved as a native `a:gradFill`; `gradientField` is its older name) | pptxgenjs gradient fill options are dropped |
 | `margin: 0` when text must align with a shape edge | boxes carry internal padding |
 | Stacked bar/column `dataLabelPosition` in `ctr`, `inEnd`, `inBase` | `outEnd` corrupts the file → runtime `chart_stacked_label_position` |
 | A secondary-axis combo needs `valAxes` and `catAxes` with two entries each | PowerPoint drops the chart → runtime `chart_axis_undeclared` |

package/src/defaults/skills/pptx/references/pictures.md

@@ -26,7 +26,7 @@ The cover decision follows the subject and available authorized assets (`directi
 Three families; the modifiers in §3 and the lenses in `composition.md` §2 do the rest. Frame ratios are what `aspect:` gets when the picture is generated.
 
 **Reference — what the frontier picture page is** (24 picture pages among 400 across thirteen decks, September 2026): twenty of them are *artifacts in a row* — three or four product screenshots, gameplay stills, press clippings, or document excerpts on one baseline across the body, each in its frame (a phone, a browser bar, a hairline) with a caption under it (Spotify's four phones on the green field, Krafton's stills grid, Bond's side-by-side captures, Evans's clipping collage); the picture covers 15-30% of the page and the page reads at 0.5-0.9 ink because the captions carry the argument. A full-bleed photograph under a scrim appears on covers and section marks, almost never on a body page. So a picture page is an evidence page: `tiles()` for the row, `clippings()` for the collage, a caption per artifact that says what it proves, and the title that says what the row shows together.
-**Reference — how the frontier photograph sits and is toned** (tile reads of the same 24 picture pages plus the Samsung, NVIDIA, Sequoia, Coatue, and Kakao decks, `picture-measure.mjs`, September 2026): the photographic region covers a median 0.25 of the canvas on a picture page (0.05-0.94; the artifact rows 0.5-0.7 of their box is actual picture, the rest is frame and caption), and sits as a *band* across the width (Samsung: the top 28-44% of every body page, fading into the paper over its lower third — luminance ramp 0.6-0.7; Bond and Coatue: the body band under the title), or as an *inset* on one side (left 9 / right 5 of 24); a full-bleed photograph appears only on a cover or section mark (Kakao p4, Krafton p23: 0.9 of the canvas). Text over the photograph is 0.00-0.02 of the picture's tiles — copy sits beside the picture or on a scrim, never on the image. The photographs are toned, not raw: saturation 0.16-0.46 (median 0.31) and luminance 0.35-0.55 in the picture tiles, where a stock photograph reads 0.6-0.8 — a muted, mid-dark image that native type can sit beside. A gradient with one shape reads 0.01-0.12 photographic and 0.17-0.5 fill: a coloured plane where the reference has a photograph. So the picture is a real photograph (supplied, or generated per §0 in one image language), toned with `treat()`, placed as a band with `fade()` or an inset, with the copy beside it.
+**Reference — how the frontier photograph sits and is toned** (tile reads of the same 24 picture pages plus the Samsung, NVIDIA, Sequoia, Coatue, and Kakao decks, September 2026): the photographic region covers a median 0.25 of the canvas on a picture page (0.05-0.94; the artifact rows 0.5-0.7 of their box is actual picture, the rest is frame and caption), and sits as a *band* across the width (Samsung: the top 28-44% of every body page, fading into the paper over its lower third — luminance ramp 0.6-0.7; Bond and Coatue: the body band under the title), or as an *inset* on one side (left 9 / right 5 of 24); a full-bleed photograph appears only on a cover or section mark (Kakao p4, Krafton p23: 0.9 of the canvas). Text over the photograph is 0.00-0.02 of the picture's tiles — copy sits beside the picture or on a scrim, never on the image. The photographs are toned, not raw: saturation 0.16-0.46 (median 0.31) and luminance 0.35-0.55 in the picture tiles, where a stock photograph reads 0.6-0.8 — a muted, mid-dark image that native type can sit beside. A gradient with one shape reads 0.01-0.12 photographic and 0.17-0.5 fill: a coloured plane where the reference has a photograph. So the picture is a real photograph (supplied, or generated per §0 in one image language), toned with `treat()`, placed as a band with `fade()` or an inset, with the copy beside it.
 
 | Family | Variants and starting geometry | Frame |
 |---|---|---|

package/src/defaults/skills/pptx/scripts/review-deck.mjs

@@ -1,4 +1,4 @@
-import { mkdir, mkdtemp, open, readFile, stat, writeFile } from 'node:fs/promises';
+import { mkdir, mkdtemp, open, readFile, rm, stat, writeFile } from 'node:fs/promises';
 import { createHash } from 'node:crypto';
 import { dirname, extname, join, resolve } from 'node:path';
 import { pathToFileURL } from 'node:url';
@@ -140,9 +140,18 @@ export async function runReview({ input, output, provider, model, effort = 'high
   const target = resolve(output);
   await mkdir(dirname(target), { recursive: true });
   const directory = await mkdtemp(join(dirname(target), 'review-input-'));
-  const { packet, files } = await preparePacket(input, directory);
-  const { runHeadlessExec } = execute ? {} : await import('../../../../headless-exec.mjs');
-  const reportFile = await open(target, 'wx');
+  let packet;
+  let files;
+  let runHeadlessExec;
+  let reportFile;
+  try {
+    ({ packet, files } = await preparePacket(input, directory));
+    ({ runHeadlessExec } = execute ? {} : await import('../../../../headless-exec.mjs'));
+    reportFile = await open(target, 'wx');
+  } catch (error) {
+    await rm(directory, { recursive: true, force: true });
+    throw error;
+  }
   let raw = '';
   const errors = [];
   const result = {

package/src/defaults/skills/skill-creator/scripts/validate-skill.mjs

@@ -34,42 +34,59 @@ function referencedPaths(body) {
   return [...found];
 }
 
-export function validateSkillDirectory(inputPath) {
-  const errors = [];
-  const warnings = [];
-  if (!inputPath) {
-    return { ok: false, errors: ['A skill directory path is required.'], warnings };
-  }
+// Resolve, read and parse SKILL.md; a failure is returned as the finished
+// validation result.
+function loadSkillDocument(inputPath) {
+  const fail = (message) => ({ failure: { ok: false, errors: [message], warnings: [] } });
+  if (!inputPath) return fail('A skill directory path is required.');
   const skillDir = resolve(String(inputPath));
   const skillFile = resolve(skillDir, 'SKILL.md');
   if (!existsSync(skillDir) || !statSync(skillDir).isDirectory()) {
-    return { ok: false, errors: [`Skill directory not found: ${skillDir}`], warnings };
+    return fail(`Skill directory not found: ${skillDir}`);
   }
   if (!existsSync(skillFile) || !statSync(skillFile).isFile()) {
-    return { ok: false, errors: [`SKILL.md not found: ${skillFile}`], warnings };
+    return fail(`SKILL.md not found: ${skillFile}`);
   }
-
   let source = '';
   try {
     source = readFileSync(skillFile, 'utf8');
   } catch (error) {
-    return {
-      ok: false,
-      errors: [`Could not read SKILL.md: ${error instanceof Error ? error.message : String(error)}`],
-      warnings,
-    };
+    return fail(`Could not read SKILL.md: ${error instanceof Error ? error.message : String(error)}`);
   }
-
-  let parsed;
   try {
-    parsed = parseSkillDocument(source);
+    return { skillDir, parsed: parseSkillDocument(source) };
   } catch (error) {
-    return {
-      ok: false,
-      errors: [error instanceof Error ? error.message : String(error)],
-      warnings,
-    };
+    return fail(error instanceof Error ? error.message : String(error));
   }
+}
+
+function checkReferencedResources(skillDir, body, errors, warnings) {
+  for (const resource of referencedPaths(body)) {
+    const resourcePath = resolve(skillDir, resource);
+    const rel = relative(skillDir, resourcePath);
+    if (!rel || rel.startsWith('..') || isAbsolute(rel)) {
+      errors.push(`Referenced resource escapes the skill directory: ${resource}.`);
+    } else if (!existsSync(resourcePath)) {
+      // A skill that bundles no folder of that kind is usually naming a
+      // repository path (`scripts/build.js` of the target repo), which this
+      // validator cannot resolve; a missing file inside a bundled folder is
+      // a broken link.
+      const bundledKind = existsSync(resolve(skillDir, resource.split('/')[0]));
+      if (bundledKind) errors.push(`Referenced resource does not exist: ${resource}.`);
+      else
+        warnings.push(
+          `${resource} is not bundled with the skill; if it is a repository path, prefer a repo-relative form the reader can resolve, or bundle it.`
+        );
+    }
+  }
+}
+
+export function validateSkillDirectory(inputPath) {
+  const loaded = loadSkillDocument(inputPath);
+  if (loaded.failure) return loaded.failure;
+  const { skillDir, parsed } = loaded;
+  const errors = [];
+  const warnings = [];
 
   const { name, description, whenToUse, body, frontmatter } = parsed;
   for (const key of Object.keys(frontmatter)) {
@@ -133,24 +150,7 @@ export function validateSkillDirectory(inputPath) {
     warnings.push(`SKILL.md body has ${bodyLines} lines; progressive disclosure is recommended above 500.`);
   }
 
-  for (const resource of referencedPaths(body)) {
-    const resourcePath = resolve(skillDir, resource);
-    const rel = relative(skillDir, resourcePath);
-    if (!rel || rel.startsWith('..') || isAbsolute(rel)) {
-      errors.push(`Referenced resource escapes the skill directory: ${resource}.`);
-    } else if (!existsSync(resourcePath)) {
-      // A skill that bundles no folder of that kind is usually naming a
-      // repository path (`scripts/build.js` of the target repo), which this
-      // validator cannot resolve; a missing file inside a bundled folder is
-      // a broken link.
-      const bundledKind = existsSync(resolve(skillDir, resource.split('/')[0]));
-      if (bundledKind) errors.push(`Referenced resource does not exist: ${resource}.`);
-      else
-        warnings.push(
-          `${resource} is not bundled with the skill; if it is a repository path, prefer a repo-relative form the reader can resolve, or bundle it.`
-        );
-    }
-  }
+  checkReferencedResources(skillDir, body, errors, warnings);
 
   return {
     ok: errors.length === 0,

package/src/defaults/skills/xlsx/SKILL.md

@@ -37,6 +37,7 @@ Excel proves the numbers and makes their meaning readable. Keep sourced inputs a
 1. Settle the table first: headers, one row per record, consistent units, numeric cells as numbers (not text), dates as ISO strings (both backends store `2026-09-30` as the Excel date under `yyyy-mm-dd`, so date arithmetic is `=D5-DATE(2026,9,22)`, not `DATEVALUE`), percentages as fractions (`0.15`, formatted `0.0%`). A figure written as text is stored as the number Excel types from it, on both backends — `"1,234"` under `#,##0`, `"15%"` as 0.15 under `0%` — and any other text stays exactly as written (`"1-2"`, `"007"`, `"1e3"`, `"TRUE"` never become a date, a 7, a 1000, or a truth value). Keep the same `design.content` model as the deck or document in the same package.
 2. For a report, design the reading order, sheet roles, column proportions, chart placement and intended screen/print use. Author with native cell, style, merge and chart operations by default. `compose_sheet` is an opt-in preset only when its table-plus-panel structure already fits; do not select it merely to save calls. Batch all known operations. Use `describe format:'xlsx' operation:<op>` when fields are unknown; cell styles belong in `properties`.
 3. Add sheets with `add_sheet`, then native ranges and formulas targeting `sheet`. Use `set_formula` for derived cells and `define_name` for reused inputs. Check representative formulas against the intended ranges before extending a model; clean recalculation proves evaluation, not correct modelling. A design specimen is optional, not required for every workbook.
+   If `create` fails, stop dependent `open`, `render`, and `finalize` calls: a failed creation is not an artifact. Correct the reported inputs and resume only from a successful `create` result.
 4. Raw tabular files (`.csv`, `.tsv`) take `set_range` with a 2D `values` array and `append_row`; they carry no styles, formulas, or multiple sheets.
 5. `finalize` recalculates and returns current rendered pages for review; add `auditProfile:'financial-model'` for a decision model. Read the actual images as in `report-design.md`, with concrete keep/fix observations about hierarchy, charts and print layout. Correct material issues and rerender, rather than filling a pass checklist. Complete with `design:{ reviewed:true, reviewToken:<current token>, critique:[{ page:1, verdict:'pass', note:<page-specific observations> }, ...] }`. This records agent review, not user approval. CSV/TSV needs no visual approval. Resolve validation failures against their original baseline, never by reopening as a new baseline.
 

package/src/headless-exec.mjs

@@ -6,10 +6,11 @@ import {
   createPristineExecutionBoundary,
   validateExplicitPristineRoute,
 } from './runtime/shared/pristine-execution.mjs';
+import { BUNDLED_PATCH_MANIFEST_PATH } from './runtime/agent/orchestrator/tools/patch-manifest-path.mjs';
 import { hasActiveBackgroundTasks } from './runtime/shared/background-tasks.mjs';
 import { installProcessSignalCleanup } from './runtime/shared/process-shutdown.mjs';
 import { closeUsageLedgers } from './runtime/shared/llm/usage-ledger.mjs';
-import { stopStandaloneMemoryRuntimesForProcess } from './standalone/memory-runtime-proxy.mjs';
+import { stopStandaloneMemoryRuntimesForProcess } from './session-runtime/services/memory-runtime-proxy.mjs';
 import { shutdownDaemonForRuntimeRoot } from './standalone/session-client.mjs';
 import { applyUsageDelta, createSessionStats } from './ui/session-stats.mjs';
 import { clean } from './runtime/shared/clean.mjs';
@@ -68,7 +69,8 @@ export async function runHeadlessExec({
   write = (text) => stdout.write(text),
   writeErr = (text) => stderr.write(text),
   usageLogPath = process.env.MIXDOG_USAGE_LOG,
-  boundaryFactory = createPristineExecutionBoundary,
+  boundaryFactory = (route) =>
+    createPristineExecutionBoundary({ ...route, patchManifestPath: BUNDLED_PATCH_MANIFEST_PATH }),
   runtimeFactory = null,
   memoryRuntimeCleanup = stopStandaloneMemoryRuntimesForProcess,
   daemonRuntimeCleanup = shutdownDaemonForRuntimeRoot,