fix(agents,skills): keep session-identity and experience-review transcript truncation UTF-16 safe

Monkey-wusky · Monkey-wusky · commit a3dd35108259 · 2026-07-16T11:01:11.000+08:00
normalizeCliContextValue (cli-output.ts) and formatSkillExperienceReviewTranscript (experience-review-prompt.ts) both used raw .slice(0, N) / .slice(-N) to truncate user-facing strings. When a truncation boundary falls inside a surrogate pair (emoji), the resulting string contains a lone surrogate — encodeURIComponent throws URIError, JSON serialisers may reject it, and downstream display surfaces can render garbage. Replace .slice(0, 200) with truncateUtf16Safe in normalizeCliContextValue. Replace .slice(0, 6_000) with truncateUtf16Safe and full.slice(-tailBudget) with sliceUtf16Safe(full, tailStart) in formatSkillExperienceReviewTranscript. Follows the pattern established by #98644, #102470, #101728 and the broader UTF-16 safety sweep already applied across the codebase.
diff --git a/src/agents/cli-output.test.ts b/src/agents/cli-output.test.ts
@@ -2592,4 +2592,33 @@ describe("createCliJsonlStreamingParser", () => {
     expect(commentaryTexts).toEqual(["Reading the file now.", "Now searching."]);
   });
 });
+
+function hasDanglingSurrogate(value: string): boolean {
+  return /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/u.test(value);
+}
+
+describe("formatCliOutputError", () => {
+  it("keeps session identity truncation UTF-16 safe near emoji boundaries", () => {
+    const result = parseCliJsonl(
+      JSON.stringify({
+        type: "result",
+        session_id: "s".repeat(199) + "😀tail",
+        terminal_reason: "max_turns",
+      }),
+      {
+        command: "claude",
+        output: "jsonl",
+        sessionIdFields: ["session_id"],
+      },
+      "claude-cli",
+    );
+
+    const error = formatCliOutputError(result!, {
+      runId: "r".repeat(199) + "🎉extra",
+      sessionId: "openclaw-" + "o".repeat(199) + "🦞end",
+    });
+
+    expect(hasDanglingSurrogate(error)).toBe(false);
+  });
+});
 /* oxlint-disable max-lines -- TODO: split this grandfathered oversized file. */
diff --git a/src/agents/cli-output.ts b/src/agents/cli-output.ts
@@ -5,6 +5,7 @@
  */
 import { normalizeLowercaseStringOrEmpty } from "@openclaw/normalization-core/string-coerce";
 import { normalizeStringEntries } from "@openclaw/normalization-core/string-normalization";
+import { truncateUtf16Safe } from "@openclaw/normalization-core/utf16-slice";
 import type { CliBackendConfig } from "../config/types.js";
 import { extractBalancedJsonFragments } from "../shared/balanced-json.js";
 import { isRecord } from "../utils.js";
@@ -62,7 +63,7 @@ export type CliOutput = {
 
 function normalizeCliContextValue(value: string | undefined): string | undefined {
   const normalized = value?.trim().replace(/\s+/g, " ");
-  return normalized ? normalized.slice(0, 200) : undefined;
+  return normalized ? truncateUtf16Safe(normalized, 200) : undefined;
 }
 
 export function formatCliOutputError(
diff --git a/src/skills/workshop/experience-review-prompt.ts b/src/skills/workshop/experience-review-prompt.ts
@@ -1,3 +1,5 @@
+import { sliceUtf16Safe, truncateUtf16Safe } from "@openclaw/normalization-core/utf16-slice";
+
 const EXPERIENCE_REVIEW_MAX_TRANSCRIPT_CHARS = 60_000;
 
 type ExperienceReviewPromptCandidate = {
@@ -61,9 +63,10 @@ export function formatSkillExperienceReviewTranscript(messages: readonly unknown
   if (full.length <= EXPERIENCE_REVIEW_MAX_TRANSCRIPT_CHARS) {
     return full;
   }
-  const first = rendered[0]?.slice(0, 6_000) ?? "";
+  const first = rendered[0] ? truncateUtf16Safe(rendered[0], 6_000) : "";
   const tailBudget = EXPERIENCE_REVIEW_MAX_TRANSCRIPT_CHARS - first.length - 80;
-  return `${first}\n\n[older trajectory omitted]\n\n${full.slice(-tailBudget)}`;
+  const tailStart = Math.max(0, full.length - tailBudget);
+  return `${first}\n\n[older trajectory omitted]\n\n${sliceUtf16Safe(full, tailStart)}`;
 }
 
 export function buildSkillExperienceReviewPrompt(
diff --git a/src/skills/workshop/experience-review.test.ts b/src/skills/workshop/experience-review.test.ts
@@ -341,3 +341,32 @@ describe("skill experience review scheduler", () => {
     expect(prompt).toContain("[tool call: exec]");
   });
 });
+
+function hasDanglingSurrogate(value: string): boolean {
+  return /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/u.test(value);
+}
+
+describe("formatSkillExperienceReviewTranscript", () => {
+  it("keeps first-message truncation UTF-16 safe at the 6 000-char boundary", () => {
+    // [user]\n is 7 chars; 5 992 padding chars reach position 5 999 where 😀 starts.
+    // Without UTF-16 safety, .slice(0, 6_000) would split the surrogate pair.
+    const content = "a".repeat(5_992) + "😀rest";
+    const messages = [{ role: "user", content }];
+    const transcript = formatSkillExperienceReviewTranscript(messages);
+
+    expect(hasDanglingSurrogate(transcript)).toBe(false);
+  });
+
+  it("keeps tail-end truncation UTF-16 safe for transcripts exceeding the limit", () => {
+    // Two long messages to trigger tail truncation (total > 60 000 chars).
+    const content = "b".repeat(35_000);
+    const messages = [
+      { role: "user", content: "a".repeat(5_992) + "🎉more" },
+      { role: "user", content: content + "🦞after" },
+    ];
+    const transcript = formatSkillExperienceReviewTranscript(messages);
+
+    expect(transcript.length).toBeLessThan(70_000);
+    expect(hasDanglingSurrogate(transcript)).toBe(false);
+  });
+});
