fix(skills): keep history scans UTF-16 safe

Leon-SK668 · Leon-SK668 · commit 917f5f2d001e · 2026-07-15T02:43:59.000+08:00
diff --git a/src/skills/workshop/history-scan-transcript-content.ts b/src/skills/workshop/history-scan-transcript-content.ts
@@ -1,3 +1,4 @@
+import { sliceUtf16Safe, truncateUtf16Safe } from "@openclaw/normalization-core/utf16-slice";
 import { filterHeartbeatTranscriptTurns } from "../../auto-reply/heartbeat-transcript-turns.js";
 import { redactSensitiveText } from "../../logging/redact.js";
 import { formatSkillExperienceReviewTranscript } from "./experience-review-prompt.js";
@@ -20,12 +21,12 @@ function capSessionTranscript(transcript: string, maxChars: number): string {
   }
   const omission = "\n\n[older session content omitted]\n\n";
   if (maxChars <= omission.length) {
-    return transcript.slice(0, maxChars);
+    return truncateUtf16Safe(transcript, maxChars);
   }
   const contentBudget = Math.max(0, maxChars - omission.length);
   const headLength = Math.min(2_000, Math.floor(contentBudget / 2));
-  const head = transcript.slice(0, headLength);
-  const tail = transcript.slice(-(contentBudget - headLength));
+  const head = truncateUtf16Safe(transcript, headLength);
+  const tail = sliceUtf16Safe(transcript, -(contentBudget - head.length));
   return `${head}${omission}${tail}`;
 }
 
diff --git a/src/skills/workshop/history-scan.test.ts b/src/skills/workshop/history-scan.test.ts
@@ -48,6 +48,10 @@ function summary(sessionKey: string, overrides: Partial<SessionEntrySummary["ent
   >;
 }
 
+function hasDanglingSurrogate(value: string): boolean {
+  return /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/u.test(value);
+}
+
 describe("Skill Workshop history scan", () => {
   it("keeps interactive sessions and excludes system-owned work", () => {
     expect(isSkillHistoryScanSessionEligible(summary("agent:main:main"))).toBe(true);
@@ -244,6 +248,33 @@ describe("Skill Workshop history scan", () => {
     expect(transcript).toContain("…redacted…");
   });
 
+  it.each([
+    {
+      name: "a prefix-only budget",
+      content: "😀tail",
+      maxChars: 8,
+      includesOmission: false,
+    },
+    {
+      name: "the retained head",
+      content: `😀${"a".repeat(80)}`,
+      maxChars: 51,
+      includesOmission: true,
+    },
+    {
+      name: "the retained tail",
+      content: `${"a".repeat(60)}😀${"x".repeat(7)}`,
+      maxChars: 51,
+      includesOmission: true,
+    },
+  ])("keeps $name UTF-16 safe", ({ content, maxChars, includesOmission }) => {
+    const transcript = formatSkillHistoryScanTranscript([{ role: "user", content }], maxChars);
+
+    expect(transcript.length).toBeLessThanOrEqual(maxChars);
+    expect(transcript.includes("[older session content omitted]")).toBe(includesOmission);
+    expect(hasDanglingSurrogate(transcript)).toBe(false);
+  });
+
   it("continues before the oldest cursor and can return to new work", () => {
     const candidates = [
       { instanceId: "new", sessionKey: "main", updatedAtMs: 300, entry: summary("new").entry },

Original file line number	Diff line number	Diff line change
`@@ -1,3 +1,4 @@`
	`1`	`+import { sliceUtf16Safe, truncateUtf16Safe } from "@openclaw/normalization-core/utf16-slice";`
`1`	`2`	`import { filterHeartbeatTranscriptTurns } from "../../auto-reply/heartbeat-transcript-turns.js";`
`2`	`3`	`import { redactSensitiveText } from "../../logging/redact.js";`
`3`	`4`	`import { formatSkillExperienceReviewTranscript } from "./experience-review-prompt.js";`
`@@ -20,12 +21,12 @@ function capSessionTranscript(transcript: string, maxChars: number): string {`
`20`	`21`	`}`
`21`	`22`	`const omission = "\n\n[older session content omitted]\n\n";`
`22`	`23`	`if (maxChars <= omission.length) {`
`23`		`- return transcript.slice(0, maxChars);`
	`24`	`+ return truncateUtf16Safe(transcript, maxChars);`
`24`	`25`	`}`
`25`	`26`	`const contentBudget = Math.max(0, maxChars - omission.length);`
`26`	`27`	`const headLength = Math.min(2_000, Math.floor(contentBudget / 2));`
`27`		`- const head = transcript.slice(0, headLength);`
`28`		`- const tail = transcript.slice(-(contentBudget - headLength));`
	`28`	`+ const head = truncateUtf16Safe(transcript, headLength);`
	`29`	`+ const tail = sliceUtf16Safe(transcript, -(contentBudget - head.length));`
`29`	`30`	return `${head}${omission}${tail}`;
`30`	`31`	`}`
`31`	`32`