ultra11y 5.31.0 → 5.31.2

package/.claude-plugin/marketplace.json

@@ -10,7 +10,7 @@
       "name": "ultra11y",
       "source": "./",
       "description": "Automatic accessibility review of the code under change before every commit, push and pull request, plus full WCAG 2.2 AA / RGAA audits and conformance deliverables on demand.",
-      "version": "5.31.0",
+      "version": "5.31.2",
       "author": { "name": "maxgfr" },
       "homepage": "https://github.com/maxgfr/ultra11y",
       "repository": "https://github.com/maxgfr/ultra11y",

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "ultra11y",
   "description": "Accessibility for WCAG 2.2 AA and country standards (RGAA…): an automatic review of the code under change before every commit, push and pull request, plus full audits and conformance deliverables on demand.",
-  "version": "5.31.0",
+  "version": "5.31.2",
   "author": { "name": "maxgfr" },
   "homepage": "https://github.com/maxgfr/ultra11y",
   "repository": "https://github.com/maxgfr/ultra11y",

package/.codex-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "ultra11y",
   "description": "Accessibility for WCAG 2.2 AA and country standards (RGAA…): an automatic review of the code under change before every commit, push and pull request, plus full audits and conformance deliverables on demand.",
-  "version": "5.31.0",
+  "version": "5.31.2",
   "author": { "name": "maxgfr", "url": "https://github.com/maxgfr" },
   "homepage": "https://github.com/maxgfr/ultra11y",
   "repository": "https://github.com/maxgfr/ultra11y",

package/.opencode/plugins/ultra11y.js

@@ -27,7 +27,7 @@ import { fileURLToPath } from "node:url";
 // Ownership marker: `ultra11y install --opencode` refuses to overwrite a plugin file that
 // does not carry it, so a user's own ultra11y.js is never clobbered.
 const ULTRA11Y_OPENCODE_PLUGIN = "ULTRA11Y_OPENCODE_PLUGIN";
-const ULTRA11Y_PLUGIN_VERSION = "5.31.0";
+const ULTRA11Y_PLUGIN_VERSION = "5.31.2";
 
 const here = dirname(fileURLToPath(import.meta.url));
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ultra11y",
-  "version": "5.31.0",
+  "version": "5.31.2",
   "description": "Audit HTML/CSS/JSX against WCAG 2.2 AA and produce a dated compliance report, or author/review accessible markup (native-HTML-first). A skills.sh agent skill: an install-free static engine (bundled, no npm install) plus the agent's judgment, with check/verify gates against hallucinated non-conformities, and pluggable in-repo standards packs (RGAA France first; Section 508 & EN 301 549 next).",
   "type": "module",
   "packageManager": "pnpm@10.33.0",

package/scripts/ultra11y.mjs

@@ -6,7 +6,7 @@ import { join as join52, relative as relative5, resolve as resolve16, sep as sep
 import { fileURLToPath as fileURLToPath5, pathToFileURL as pathToFileURL3 } from "url";
 
 // src/types.ts
-var VERSION = "5.31.0";
+var VERSION = "5.31.2";
 var SCHEMA_VERSION = 2;
 var INAPPLICABLE_STATUS = "C";
 
@@ -60209,7 +60209,9 @@ async function judgeAll(batches, opts) {
     Array.from({ length: Math.min(lanes, queue.length) }, async () => {
       for (let b = queue.shift(); b !== void 0; b = queue.shift()) {
         try {
-          verdicts.push(...await backend(b.items, b.prompt, opts));
+          const landed = await backend(b.items, b.prompt, opts);
+          verdicts.push(...landed);
+          if (landed.length) opts.onVerdicts?.(landed);
         } catch (e) {
           failures.push(e instanceof Error ? e.message : String(e));
         }
@@ -68733,6 +68735,11 @@ function isCommand(s) {
   return !!s && COMMANDS.includes(s);
 }
 var VALUE_FLAGS2 = /* @__PURE__ */ new Set([
+  "runner",
+  "grain",
+  "max-budget-usd",
+  "timeout",
+  "concurrency",
   "out",
   "provider",
   "grain",
@@ -70720,6 +70727,29 @@ async function cmdJudge(p) {
   console.error(
     lang === "fr" ? `ultra11y judge : ${items.length} crit\xE8re(s) en ${batches.length} lot(s), transport ${runner}, mod\xE8le ${model}\u2026` : `ultra11y judge: ${items.length} criterion(ia) in ${batches.length} batch(es), ${runner} transport, model ${model}\u2026`
   );
+  const outDir = typeof p.flags.out === "string" ? p.flags.out : ".";
+  const cwd = typeof p.flags.cwd === "string" ? p.flags.cwd : void 0;
+  const applying = p.flags.apply === true;
+  const askedConcurrency = typeof p.flags.concurrency === "string" ? Number(p.flags.concurrency) : Number.NaN;
+  const cliConcurrency = Number.isFinite(askedConcurrency) && askedConcurrency > 0 ? Math.min(askedConcurrency, 8) : 2;
+  const CHECKPOINT_MS = 15e3;
+  let lastCheckpoint = 0;
+  const checkpoint = (force) => {
+    if (!applying) return;
+    const now = Date.now();
+    if (!force && now - lastCheckpoint < CHECKPOINT_MS) return;
+    lastCheckpoint = now;
+    try {
+      const partial = applyAdjudication(
+        audit2,
+        { tool: "ultra11y", kind: "adjudication", schemaVersion: SCHEMA_VERSION, standard, auditDate: audit2.date, items },
+        { cwd }
+      );
+      if (partial.applied > 0) writeFileSync20(join52(outDir, "audit-latest.json"), `${JSON.stringify(partial.audit, null, 2)}
+`);
+    } catch {
+    }
+  };
   let spent = 0;
   const maxBudgetUsd = typeof p.flags["max-budget-usd"] === "string" ? Number(p.flags["max-budget-usd"]) : void 0;
   const timeoutMs = typeof p.flags.timeout === "string" ? Number(p.flags.timeout) * 1e3 : void 0;
@@ -70727,14 +70757,22 @@ async function cmdJudge(p) {
     apiKey: key,
     model,
     backend: runner === "cli" ? judgeBatchCli : void 0,
-    // One local process at a time. Four in flight is an answer to a rate limit, and a rate
-    // limit is not what bounds a subprocess.
-    concurrency: runner === "cli" ? 1 : void 0,
+    // TWO local processes, not one, and the reason is measured rather than tuned: sequentially,
+    // one criterion took ~75s on Haiku, so a 51-criterion sweep needs ~64 minutes and a CI job
+    // that allows 45 is killed at 31. Two in flight brings it inside the ceiling; more than a
+    // few is a different risk — each `claude` is its own Node process making its own API calls,
+    // and four of them on a subscription token reach a rate limit faster than four HTTP
+    // requests do. `--concurrency` overrides it for a runner with room.
+    concurrency: runner === "cli" ? cliConcurrency : void 0,
     maxBudgetUsd: Number.isFinite(maxBudgetUsd) ? maxBudgetUsd : void 0,
     timeoutMs: Number.isFinite(timeoutMs) ? timeoutMs : void 0,
     onCost: (usd) => {
       spent += usd;
     },
+    onVerdicts: (landed) => {
+      applyRawVerdicts(items, landed);
+      checkpoint(false);
+    },
     onProgress: (done, total) => console.error(`  ${done}/${total}`)
   });
   if (spent > 0) console.error(lang === "fr" ? `ultra11y judge : ${spent.toFixed(4)} $ d\xE9pens\xE9(s).` : `ultra11y judge: ${spent.toFixed(4)} spent.`);

package/skills/review-a11y/SKILL.md

@@ -5,7 +5,7 @@ when_to_use: "Invoke WITHOUT being asked whenever changed front-end code should
 allowed-tools: Bash(node ${CLAUDE_SKILL_DIR}/scripts/ultra11y.mjs *), Read, Grep, Glob
 license: MIT
 metadata:
-  version: 5.31.0
+  version: 5.31.2
 ---
 
 # review-a11y — accessibility review of changed code

package/skills/review-a11y/scripts/ultra11y.mjs

@@ -6,7 +6,7 @@ import { join as join52, relative as relative5, resolve as resolve16, sep as sep
 import { fileURLToPath as fileURLToPath5, pathToFileURL as pathToFileURL3 } from "url";
 
 // src/types.ts
-var VERSION = "5.31.0";
+var VERSION = "5.31.2";
 var SCHEMA_VERSION = 2;
 var INAPPLICABLE_STATUS = "C";
 
@@ -60209,7 +60209,9 @@ async function judgeAll(batches, opts) {
     Array.from({ length: Math.min(lanes, queue.length) }, async () => {
       for (let b = queue.shift(); b !== void 0; b = queue.shift()) {
         try {
-          verdicts.push(...await backend(b.items, b.prompt, opts));
+          const landed = await backend(b.items, b.prompt, opts);
+          verdicts.push(...landed);
+          if (landed.length) opts.onVerdicts?.(landed);
         } catch (e) {
           failures.push(e instanceof Error ? e.message : String(e));
         }
@@ -68733,6 +68735,11 @@ function isCommand(s) {
   return !!s && COMMANDS.includes(s);
 }
 var VALUE_FLAGS2 = /* @__PURE__ */ new Set([
+  "runner",
+  "grain",
+  "max-budget-usd",
+  "timeout",
+  "concurrency",
   "out",
   "provider",
   "grain",
@@ -70720,6 +70727,29 @@ async function cmdJudge(p) {
   console.error(
     lang === "fr" ? `ultra11y judge : ${items.length} crit\xE8re(s) en ${batches.length} lot(s), transport ${runner}, mod\xE8le ${model}\u2026` : `ultra11y judge: ${items.length} criterion(ia) in ${batches.length} batch(es), ${runner} transport, model ${model}\u2026`
   );
+  const outDir = typeof p.flags.out === "string" ? p.flags.out : ".";
+  const cwd = typeof p.flags.cwd === "string" ? p.flags.cwd : void 0;
+  const applying = p.flags.apply === true;
+  const askedConcurrency = typeof p.flags.concurrency === "string" ? Number(p.flags.concurrency) : Number.NaN;
+  const cliConcurrency = Number.isFinite(askedConcurrency) && askedConcurrency > 0 ? Math.min(askedConcurrency, 8) : 2;
+  const CHECKPOINT_MS = 15e3;
+  let lastCheckpoint = 0;
+  const checkpoint = (force) => {
+    if (!applying) return;
+    const now = Date.now();
+    if (!force && now - lastCheckpoint < CHECKPOINT_MS) return;
+    lastCheckpoint = now;
+    try {
+      const partial = applyAdjudication(
+        audit2,
+        { tool: "ultra11y", kind: "adjudication", schemaVersion: SCHEMA_VERSION, standard, auditDate: audit2.date, items },
+        { cwd }
+      );
+      if (partial.applied > 0) writeFileSync20(join52(outDir, "audit-latest.json"), `${JSON.stringify(partial.audit, null, 2)}
+`);
+    } catch {
+    }
+  };
   let spent = 0;
   const maxBudgetUsd = typeof p.flags["max-budget-usd"] === "string" ? Number(p.flags["max-budget-usd"]) : void 0;
   const timeoutMs = typeof p.flags.timeout === "string" ? Number(p.flags.timeout) * 1e3 : void 0;
@@ -70727,14 +70757,22 @@ async function cmdJudge(p) {
     apiKey: key,
     model,
     backend: runner === "cli" ? judgeBatchCli : void 0,
-    // One local process at a time. Four in flight is an answer to a rate limit, and a rate
-    // limit is not what bounds a subprocess.
-    concurrency: runner === "cli" ? 1 : void 0,
+    // TWO local processes, not one, and the reason is measured rather than tuned: sequentially,
+    // one criterion took ~75s on Haiku, so a 51-criterion sweep needs ~64 minutes and a CI job
+    // that allows 45 is killed at 31. Two in flight brings it inside the ceiling; more than a
+    // few is a different risk — each `claude` is its own Node process making its own API calls,
+    // and four of them on a subscription token reach a rate limit faster than four HTTP
+    // requests do. `--concurrency` overrides it for a runner with room.
+    concurrency: runner === "cli" ? cliConcurrency : void 0,
     maxBudgetUsd: Number.isFinite(maxBudgetUsd) ? maxBudgetUsd : void 0,
     timeoutMs: Number.isFinite(timeoutMs) ? timeoutMs : void 0,
     onCost: (usd) => {
       spent += usd;
     },
+    onVerdicts: (landed) => {
+      applyRawVerdicts(items, landed);
+      checkpoint(false);
+    },
     onProgress: (done, total) => console.error(`  ${done}/${total}`)
   });
   if (spent > 0) console.error(lang === "fr" ? `ultra11y judge : ${spent.toFixed(4)} $ d\xE9pens\xE9(s).` : `ultra11y judge: ${spent.toFixed(4)} spent.`);

package/skills/ultra11y/SKILL.md

@@ -4,7 +4,7 @@ description: "Use to AUDIT existing HTML/CSS/JSX against WCAG 2.2 AA accessibili
 when_to_use: "Invoke when the user asks for an accessibility AUDIT or a formal deliverable: audit a repo, site or page against WCAG 2.2 AA or a country standard (RGAA, Section 508, EN 301 549); produce a dated conformance report, a criterion grid, a PRD backlog, or tickets (GitHub/GitLab/Jira); author accessible markup from scratch; wire the repo gate (init --hook/--ci) or a rendered-DOM capture pipeline. For a review of the code UNDER CHANGE (diff, branch, PR), use the `review-a11y` skill instead."
 license: MIT
 metadata:
-  version: 5.31.0
+  version: 5.31.2
 ---
 
 # ultra11y — audit WCAG 2.2 AA and write accessible markup

package/skills/ultra11y/scripts/ultra11y.mjs

@@ -6,7 +6,7 @@ import { join as join52, relative as relative5, resolve as resolve16, sep as sep
 import { fileURLToPath as fileURLToPath5, pathToFileURL as pathToFileURL3 } from "url";
 
 // src/types.ts
-var VERSION = "5.31.0";
+var VERSION = "5.31.2";
 var SCHEMA_VERSION = 2;
 var INAPPLICABLE_STATUS = "C";
 
@@ -60209,7 +60209,9 @@ async function judgeAll(batches, opts) {
     Array.from({ length: Math.min(lanes, queue.length) }, async () => {
       for (let b = queue.shift(); b !== void 0; b = queue.shift()) {
         try {
-          verdicts.push(...await backend(b.items, b.prompt, opts));
+          const landed = await backend(b.items, b.prompt, opts);
+          verdicts.push(...landed);
+          if (landed.length) opts.onVerdicts?.(landed);
         } catch (e) {
           failures.push(e instanceof Error ? e.message : String(e));
         }
@@ -68733,6 +68735,11 @@ function isCommand(s) {
   return !!s && COMMANDS.includes(s);
 }
 var VALUE_FLAGS2 = /* @__PURE__ */ new Set([
+  "runner",
+  "grain",
+  "max-budget-usd",
+  "timeout",
+  "concurrency",
   "out",
   "provider",
   "grain",
@@ -70720,6 +70727,29 @@ async function cmdJudge(p) {
   console.error(
     lang === "fr" ? `ultra11y judge : ${items.length} crit\xE8re(s) en ${batches.length} lot(s), transport ${runner}, mod\xE8le ${model}\u2026` : `ultra11y judge: ${items.length} criterion(ia) in ${batches.length} batch(es), ${runner} transport, model ${model}\u2026`
   );
+  const outDir = typeof p.flags.out === "string" ? p.flags.out : ".";
+  const cwd = typeof p.flags.cwd === "string" ? p.flags.cwd : void 0;
+  const applying = p.flags.apply === true;
+  const askedConcurrency = typeof p.flags.concurrency === "string" ? Number(p.flags.concurrency) : Number.NaN;
+  const cliConcurrency = Number.isFinite(askedConcurrency) && askedConcurrency > 0 ? Math.min(askedConcurrency, 8) : 2;
+  const CHECKPOINT_MS = 15e3;
+  let lastCheckpoint = 0;
+  const checkpoint = (force) => {
+    if (!applying) return;
+    const now = Date.now();
+    if (!force && now - lastCheckpoint < CHECKPOINT_MS) return;
+    lastCheckpoint = now;
+    try {
+      const partial = applyAdjudication(
+        audit2,
+        { tool: "ultra11y", kind: "adjudication", schemaVersion: SCHEMA_VERSION, standard, auditDate: audit2.date, items },
+        { cwd }
+      );
+      if (partial.applied > 0) writeFileSync20(join52(outDir, "audit-latest.json"), `${JSON.stringify(partial.audit, null, 2)}
+`);
+    } catch {
+    }
+  };
   let spent = 0;
   const maxBudgetUsd = typeof p.flags["max-budget-usd"] === "string" ? Number(p.flags["max-budget-usd"]) : void 0;
   const timeoutMs = typeof p.flags.timeout === "string" ? Number(p.flags.timeout) * 1e3 : void 0;
@@ -70727,14 +70757,22 @@ async function cmdJudge(p) {
     apiKey: key,
     model,
     backend: runner === "cli" ? judgeBatchCli : void 0,
-    // One local process at a time. Four in flight is an answer to a rate limit, and a rate
-    // limit is not what bounds a subprocess.
-    concurrency: runner === "cli" ? 1 : void 0,
+    // TWO local processes, not one, and the reason is measured rather than tuned: sequentially,
+    // one criterion took ~75s on Haiku, so a 51-criterion sweep needs ~64 minutes and a CI job
+    // that allows 45 is killed at 31. Two in flight brings it inside the ceiling; more than a
+    // few is a different risk — each `claude` is its own Node process making its own API calls,
+    // and four of them on a subscription token reach a rate limit faster than four HTTP
+    // requests do. `--concurrency` overrides it for a runner with room.
+    concurrency: runner === "cli" ? cliConcurrency : void 0,
     maxBudgetUsd: Number.isFinite(maxBudgetUsd) ? maxBudgetUsd : void 0,
     timeoutMs: Number.isFinite(timeoutMs) ? timeoutMs : void 0,
     onCost: (usd) => {
       spent += usd;
     },
+    onVerdicts: (landed) => {
+      applyRawVerdicts(items, landed);
+      checkpoint(false);
+    },
     onProgress: (done, total) => console.error(`  ${done}/${total}`)
   });
   if (spent > 0) console.error(lang === "fr" ? `ultra11y judge : ${spent.toFixed(4)} $ d\xE9pens\xE9(s).` : `ultra11y judge: ${spent.toFixed(4)} spent.`);

package/src/cli.ts

@@ -2,6 +2,7 @@ import { realpathSync, writeFileSync, mkdirSync, existsSync, readFileSync, appen
 import { join, relative, resolve, sep, dirname } from "node:path";
 import { fileURLToPath, pathToFileURL } from "node:url";
 import {
+  SCHEMA_VERSION,
   VERSION,
   type Lang,
   type AuditResult,
@@ -542,6 +543,11 @@ function isCommand(s: string | undefined): s is Command {
 }
 
 const VALUE_FLAGS = new Set([
+  "runner",
+  "grain",
+  "max-budget-usd",
+  "timeout",
+  "concurrency",
   "out",
   "provider",
   "grain",
@@ -3273,6 +3279,44 @@ async function cmdJudge(p: ParsedArgs): Promise<number> {
       : `ultra11y judge: ${items.length} criterion(ia) in ${batches.length} batch(es), ${runner} transport, model ${model}…`,
   );
 
+  const outDir = typeof p.flags.out === "string" ? (p.flags.out as string) : ".";
+  const cwd = typeof p.flags.cwd === "string" ? (p.flags.cwd as string) : undefined;
+  const applying = p.flags.apply === true;
+  const askedConcurrency = typeof p.flags.concurrency === "string" ? Number(p.flags.concurrency) : Number.NaN;
+  const cliConcurrency = Number.isFinite(askedConcurrency) && askedConcurrency > 0 ? Math.min(askedConcurrency, 8) : 2;
+
+  // CHECKPOINTS, because a killed run used to lose everything it had already paid for.
+  //
+  // Measured: a per-criterion pass reached 31 of 51 criteria in 40 minutes, the CI job hit its
+  // 45-minute ceiling, the process was killed — and the audit came back with all 51 still to
+  // assess. The fold only ran at the END, so 31 rulings were bought and thrown away. Isolating
+  // the model call per criterion buys nothing while the WRITE is still one all-or-nothing
+  // operation, which is exactly what the per-criterion grain was sold as fixing.
+  //
+  // Time-throttled rather than every batch: the fold re-grounds every citation against real
+  // source, and doing that 51 times over would spend on bookkeeping what the grain saves.
+  const CHECKPOINT_MS = 15_000;
+  let lastCheckpoint = 0;
+  const checkpoint = (force: boolean): void => {
+    if (!applying) return;
+    const now = Date.now();
+    if (!force && now - lastCheckpoint < CHECKPOINT_MS) return;
+    lastCheckpoint = now;
+    try {
+      const partial = applyAdjudication(
+        audit,
+        { tool: "ultra11y", kind: "adjudication", schemaVersion: SCHEMA_VERSION, standard, auditDate: audit.date, items },
+        { cwd },
+      );
+      // A partial sweep cannot satisfy the COVERAGE check, and that is correct rather than a
+      // problem: the per-verdict fold still lands every valid verdict, and the criteria nobody
+      // has reached yet stay to assess. Only a checkpoint that landed nothing is worth skipping.
+      if (partial.applied > 0) writeFileSync(join(outDir, "audit-latest.json"), `${JSON.stringify(partial.audit, null, 2)}\n`);
+    } catch {
+      // A checkpoint is a safety net, never a reason to lose the run it protects.
+    }
+  };
+
   let spent = 0;
   // A dollar ceiling and a wall clock — never a turn budget. `--max-turns` is not a flag of
   // the Claude Code CLI, and the CLI swallows unknown flags without a word, so passing one
@@ -3283,14 +3327,22 @@ async function cmdJudge(p: ParsedArgs): Promise<number> {
     apiKey: key,
     model,
     backend: runner === "cli" ? judgeBatchCli : undefined,
-    // One local process at a time. Four in flight is an answer to a rate limit, and a rate
-    // limit is not what bounds a subprocess.
-    concurrency: runner === "cli" ? 1 : undefined,
+    // TWO local processes, not one, and the reason is measured rather than tuned: sequentially,
+    // one criterion took ~75s on Haiku, so a 51-criterion sweep needs ~64 minutes and a CI job
+    // that allows 45 is killed at 31. Two in flight brings it inside the ceiling; more than a
+    // few is a different risk — each `claude` is its own Node process making its own API calls,
+    // and four of them on a subscription token reach a rate limit faster than four HTTP
+    // requests do. `--concurrency` overrides it for a runner with room.
+    concurrency: runner === "cli" ? cliConcurrency : undefined,
     maxBudgetUsd: Number.isFinite(maxBudgetUsd) ? maxBudgetUsd : undefined,
     timeoutMs: Number.isFinite(timeoutMs) ? timeoutMs : undefined,
     onCost: (usd) => {
       spent += usd;
     },
+    onVerdicts: (landed) => {
+      applyRawVerdicts(items, landed);
+      checkpoint(false);
+    },
     onProgress: (done, total) => console.error(`  ${done}/${total}`),
   });
   // What it cost, from the run itself rather than from a dashboard read a day later.

package/src/llm.ts

@@ -74,6 +74,15 @@ export interface LlmOptions {
   /** Called with each invocation's real cost, so a run can report what it spent instead of
    *  leaving the reader to find it in a provider dashboard. */
   onCost?: (usd: number) => void;
+  /** Called with each batch's verdicts AS THEY LAND, so a caller can persist them before the
+   *  run is over.
+   *
+   *  Measured, and it cost a whole paid run: a per-criterion pass reached 31 of 51 criteria in
+   *  40 minutes, the CI job hit its 45-minute ceiling, the process was killed — and because
+   *  the fold only ran at the END, all 31 verdicts went with it. The audit came back with 51
+   *  criteria still to assess, having paid for 31. Isolating the model CALL per criterion is
+   *  worth nothing if the write is still one all-or-nothing operation at the end. */
+  onVerdicts?: (verdicts: RawVerdict[]) => void;
   /** Backoff before retry N (1-based). Injected so a test can exercise the retry path
    *  without waiting out the real curve. */
   backoffMs?: (attempt: number) => number;
@@ -270,7 +279,11 @@ export async function judgeAll(
     Array.from({ length: Math.min(lanes, queue.length) }, async () => {
       for (let b = queue.shift(); b !== undefined; b = queue.shift()) {
         try {
-          verdicts.push(...(await backend(b.items, b.prompt, opts)));
+          const landed = await backend(b.items, b.prompt, opts);
+          verdicts.push(...landed);
+          // Handed over BEFORE the next batch starts, so a run that dies mid-sweep leaves
+          // behind what it had already ruled on rather than nothing at all.
+          if (landed.length) opts.onVerdicts?.(landed);
         } catch (e) {
           failures.push(e instanceof Error ? e.message : String(e));
         }

package/src/types.ts

@@ -1,6 +1,6 @@
 // Single source of truth for shared types + the embedded version. sync-version.mjs
 // rewrites VERSION at release time (kept in lockstep with package.json + SKILL.md).
-export const VERSION = "5.31.0";
+export const VERSION = "5.31.2";
 export const SCHEMA_VERSION = 2;
 
 // Lang = the UI FRAME's language — the `L` localization tables each module (report,