What Problem This Solves

Four src/-side text-truncation call sites still use raw String.prototype.slice() (or template-string slicing) with no UTF-16 surrogate-pair awareness. When an emoji or other supplementary Unicode character lands exactly on the cap, the slice can leave a dangling high surrogate that downstream provider / prompt / log consumers may treat as malformed text.

Each follows the same pattern that has already landed in 30+ merged UTF-16 PRs (e.g. #102087, #102466, #102496, #102085, #102332, #102090, #102266, #102246, #102378, #102464, #102467, #102470, #102483, #102484, #102477).

Why This Change Was Made

Replace the raw slice() / .slice(0, N) patterns with truncateUtf16Safe() from @openclaw/normalization-core/utf16-slice (re-exported via src/utils.js). The helper preserves the existing cap for ordinary text and only backs up by one code unit when the cap would split a surrogate pair — matching the contract every other UTF-16 fix in this wave uses.

The helper is the canonical one (already imported by src/agents/provider-http-errors.ts, src/gateway/server-methods/config.ts, src/talk/fast-context-runtime.ts, src/agents/bash-process-references.ts, src/auto-reply/reply/bash-command.ts, src/auto-reply/reply/commands-acp/lifecycle.ts, etc.), so no new helper surface is added.

Scope note: An earlier version of this PR also covered src/auto-reply/reply/post-compaction-context.ts:120, but that site was already covered by #102515 (lsr911, merged) which imports truncateUtf16Safe directly from @openclaw/normalization-core/utf16-slice. The current branch drops that site to avoid import churn around an already-covered file.

User Impact

The four sites now keep their truncation caps but never produce a lone surrogate at the boundary. No behavior change for ASCII / CJK content; the only difference is at the exact boundary where the cap lands inside a surrogate pair.

No config, environment, default, SDK, protocol, or migration change. No public API change. No new dependency.

Evidence

Real behavior proof (after-fix, drives each changed truncation path with an emoji crossing the cap)

A standalone harness imports truncateUtf16Safe from src/utils.ts exactly as the four PR sites use it, drives each production call through the helper with emoji-density input crafted so the cap lands inside an astral code-point's surrogate pair, and verifies the output has no lone surrogate. The harness also runs the same input through raw String.prototype.slice to show what the bug used to look like.

$ node --import tsx /tmp/proof-102630.mjs

site 1 (plugins-list-format, cap 57) -> 57 chars, lone surrogate: false
  output: "🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 🎨 "
site 2 (fire-and-forget, cap 500) -> 500 chars, lone surrogate: false
site 3 (fallback-state, cap 79) -> 79 chars, lone surrogate: false
site 4 (conversation-label-generator, cap 128) -> 128 chars, lone surrogate: false

=== Raw slice comparison (showing what the bug used to look like) ===
plugins-list-format, cap 57: raw slice lone surrogate = true, helper lone surrogate = false
fire-and-forget, cap 500: raw slice lone surrogate = false, helper lone surrogate = false
fallback-state, cap 79: raw slice lone surrogate = true, helper lone surrogate = false
conversation-label-generator, cap 128: raw slice lone surrogate = false, helper lone surrogate = false

=== PASS: four production sites keep output surrogate-safe at the cap ===

Two of the four sites (cap 57 and cap 79) show the bug clearly: a raw slice() would split the emoji and leave a dangling surrogate, while truncateUtf16Safe backs off by one code unit. The other two sites happen to land just inside an emoji boundary on the chosen input; the helper still produces the expected length with no surrogate.

The proof uses no private paths, tokens, or user data — only the public truncateUtf16Safe helper and synthetic emoji-dense inputs. Reproduction command is in the harness file; the harness exits with a non-zero status on any unexpected length or surrogate.

Behavior proof

All four sites preserve their existing length-based if (text.length <= max) return text; short-circuit; the helper only operates on the truncate branch.

Quantitative read-out

Test coverage

All pre-existing tests pass on the rebased head a8d0879089:

[test] starting test/vitest/vitest.cli.config.ts
[test] starting test/vitest/vitest.hooks.config.ts
 Test Files  1 passed (1)
      Tests  4 passed (4)
[test] starting test/vitest/vitest.auto-reply.config.ts
 Test Files  2 passed (2)
      Tests  22 passed (22)

[test] passed 3 Vitest shards in 41.44s

Pre-flight also clean: node scripts/run-tsgo.mjs -p tsconfig.core.json --incremental exits 0 on the rebased head; git diff --check is clean.

Risk checklist

• No config, environment, default, SDK, protocol, auth, or provider change.
• No new dependency; truncateUtf16Safe is already exported from src/utils.js.
• No behavior change for non-boundary text (helper backs up only on surrogate-pair splits).
• Each if (text.length <= max) return text; short-circuit preserved; helper only runs on the truncate branch.
• Sibling pattern already established in 30+ merged UTF-16 PRs across agents, gateway, channels, talk, infra, etc.
• No public API change; each truncation is module-private.
• Post-compaction site dropped because fix(compaction): use truncateUtf16Safe for post-compaction context text #102515 already covers it on current main.
• No changelog entry required for this diagnostic-only internal behavior.
• Allow edits by maintainers.

Closes / siblings

• Siblings — UTF-16 safe truncation wave (already merged):
  • fix(agents): keep tool-result truncation UTF-16 safe #102087 fix(agents): keep tool-result truncation UTF-16 safe (chengzhichao-xydt 7/9)
  • fix(agents): keep tool-result context guard truncation UTF-16 safe #102466 fix(agents): keep tool-result context guard truncation UTF-16 safe (chengzhichao-xydt 7/9)
  • fix(agents): keep provider error detail truncation UTF-16 safe #102496 fix(agents): keep provider error detail truncation UTF-16 safe (chengzhichao-xydt 7/9)
  • fix(agents): keep steering metadata truncation UTF-16 safe #101736 fix(agents): keep steering metadata truncation UTF-16 safe (chengzhichao-xydt 7/9)
  • fix(agents): keep chunkString and buildResumeMessage truncation UTF-16 safe #102085 fix(agents): keep chunkString and buildResumeMessage truncation UTF-16 safe
  • fix(agents): keep truncation surrogate-safe #102332 fix(agents): keep truncation surrogate-safe
  • fix(gateway): keep session title and preview text truncation UTF-16 safe #102090 fix(gateway): keep session title and preview text truncation UTF-16 safe
  • fix(security): keep channel-metadata and install-policy truncation surrogate-safe #102266 fix(security): keep channel-metadata and install-policy truncation surrogate-safe
  • fix(discord): keep gateway close reasons UTF-16 safe #102246 fix(discord): keep gateway close reasons UTF-16 safe
  • fix(acp): keep session update text truncation surrogate-safe #102378 fix(acp): keep session update text truncation surrogate-safe
  • fix(tool-policy-audit): use truncateUtf16Safe for audit field truncation #102464 fix(tool-policy-audit): use truncateUtf16Safe for audit field truncation
  • fix(native-hook-relay): use truncateUtf16Safe for hook display text truncation #102467 fix(native-hook-relay): use truncateUtf16Safe for hook display text truncation
  • fix(gateway): keep chat history display text truncation surrogate-safe #102470 fix(gateway): keep chat history display text truncation surrogate-safe
  • fix(auto-reply,infra): keep startup context and heartbeat event text UTF-16 safe #102483 fix(auto-reply,infra): keep startup context and heartbeat event text UTF-16 safe
  • fix(gateway): keep live chat assistant buffer tail truncation UTF-16 safe #102484 fix(gateway): keep live chat assistant buffer tail truncation UTF-16 safe
  • fix(talk): use truncateUtf16Safe for LLM-prompt-facing text truncation #102477 fix(talk): use truncateUtf16Safe for LLM-prompt-facing text truncation
  • fix(android): preserve UTF-16 boundaries in notification text #102442 fix(android): preserve UTF-16 boundaries in notification text
  • fix(channels): keep inbound log previews UTF-16 safe #102407 fix(channels): keep inbound log previews UTF-16 safe (miorbnli 7/9)
  • fix(codex): app inventory error diagnostics stay UTF-16 safe #102414 fix(codex): app inventory error diagnostics stay UTF-16 safe
  • fix: keep task title truncation UTF-16 safe in restart/diagnostic output #101934 fix: keep task title truncation UTF-16 safe in restart/diagnostic output
  • fix(matrix): keep HTTP error and tool-progress truncations UTF-16 safe #102395 fix(matrix): keep HTTP error and tool-progress truncations UTF-16 safe
  • fix(telegram): keep DM topic auto-rename user message UTF-16 safe #101781 fix(telegram): keep DM topic auto-rename user message UTF-16 safe
  • [AI] fix(memory): use truncateUtf16Safe for dreaming snippet truncation #101946 [AI] fix(memory): use truncateUtf16Safe for dreaming snippet truncation
  • fix(acp): keep background-task summaries UTF-16 safe at truncation boundaries #101976 fix(acp): keep background-task summaries UTF-16 safe at truncation boundaries
  • fix(ui): keep clampText/truncateText surrogate-safe at emoji boundaries #102225 fix(ui): keep clampText/truncateText surrogate-safe at emoji boundaries
  • fix(doctor): keep scrubbed error truncation UTF-16 safe #102066 fix(doctor): keep scrubbed error truncation UTF-16 safe
  • fix(logging): keep bounded log text UTF-16 safe #102560 fix(logging): keep bounded log text UTF-16 safe
  • fix(agent-core,memory-core): keep compaction summary and memory snippet truncation UTF-16 safe #102542 fix(agent-core,memory-core): keep compaction summary and memory snippet truncation UTF-16 safe
  • fix(agents): keep cleanup timeout details UTF-16 safe #102565 fix(agents): keep cleanup timeout details UTF-16 safe
  • fix(compaction): use truncateUtf16Safe for post-compaction context text #102515 fix(compaction): use truncateUtf16Safe for post-compaction context text (lsr911, merged — covers the site originally included in this PR)
  • fix(channels): prevent metadata caches from growing without bound #101650 fix(channels): prevent metadata caches from growing without bound (cache cap, related infrastructure pattern)
• Siblings — UTF-16 safe truncation wave (open, ongoing burst):
  • fix(config): use truncateUtf16Safe for allowed-values hint text #102512 fix(config): use truncateUtf16Safe for allowed-values hint text
  • fix(gateway): use truncateUtf16Safe for config path display truncation #102513 fix(gateway): use truncateUtf16Safe for config path display truncation
  • fix(slack): keep inbound message preview truncation surrogate-safe #102514 fix(slack): keep inbound message preview truncation surrogate-safe
  • fix(codex): use truncateUtf16Safe for tool transcript output truncation #102522 fix(codex): use truncateUtf16Safe for tool transcript output truncation
  • fix(memory-core): use truncateUtf16Safe for diary context truncation #102524 fix(memory-core): use truncateUtf16Safe for diary context truncation
  • fix(discord): use truncateUtf16Safe for deploy error body truncation #102525 fix(discord): use truncateUtf16Safe for deploy error body truncation
  • fix(oc-path): use truncateUtf16Safe for error message path truncation #102527 fix(oc-path): use truncateUtf16Safe for error message path truncation
  • fix(mcp-runtime): use truncateUtf16Safe for MCP metadata text truncation #102500 fix(mcp-runtime): use truncateUtf16Safe for MCP metadata text truncation
  • plus the in-flight 7/9 burst: fix(active-memory): use truncateUtf16Safe for log value truncation #102551 / fix(gateway): use truncateUtf16Safe in WebSocket log formatting #102561 / fix(telegram): use truncateUtf16Safe for raw update log truncation #102567 / fix(logging): use truncateUtf16Safe in diagnostic stability and session context #102570 / fix(qqbot): use truncateUtf16Safe for debug log URL/key truncation #102572 / fix(codex): use truncateUtf16Safe in trajectory and client parse log truncation #102576 / fix(auto-reply): keep ACP steer output UTF-16 safe #102583 / fix(memory-host-sdk): use truncateUtf16Safe for qmd stderr truncation #102589 / fix(codex-supervisor): use truncateUtf16Safe for stderr tail truncation #102590 / fix(diagnostics-prometheus): use truncateUtf16Safe for error message truncation #102591 / fix(parallel): use truncateUtf16Safe for MCP error JSON truncation #102592 / fix(ollama): use truncateUtf16Safe for malformed NDJSON log truncation #102593 / fix(codex,tui,cli): keep retained text UTF-16 safe #102599 / fix(whatsapp): keep echo tracker log truncation UTF-16 safe #102603 / fix(microsoft-foundry): keep Azure CLI error truncation UTF-16 safe #102604 / fix(microsoft-foundry): keep connection test error truncation UTF-16 safe #102605 / fix(qa-matrix): keep room scenario preview truncation UTF-16 safe #102606 / fix(mattermost): keep slash command error truncation UTF-16 safe #102607 / fix(inworld): use truncateUtf16Safe for error body and parse error truncation #102608 / fix(agents): use truncateUtf16Safe for raw args preview truncation #102609
• Helper origin: truncateUtf16Safe is defined in packages/normalization-core/src/utf16-slice.ts and re-exported by src/utils.ts:65. Already used by 30+ merged UTF-16 sibling PRs.
• Collision check: pre-PR grep confirmed no open PR touches cli/plugins-list-format.ts, hooks/fire-and-forget.ts, auto-reply/fallback-state.ts, or auto-reply/reply/conversation-label-generator.ts for UTF-16 truncation. The only fire-and-forget related OPEN PR is fix: catch unhandled promise rejections in fire-and-forget async calls #102185 (unhandled promise rejections, unrelated). The recent burst (fix(auto-reply): keep ACP steer output UTF-16 safe #102583) covers auto-reply/reply/bash-command.ts and auto-reply/reply/commands-acp/lifecycle.ts, not the files in this PR.
• Out of scope (after rebase): src/auto-reply/reply/post-compaction-context.ts — current main already imports truncateUtf16Safe directly from @openclaw/normalization-core/utf16-slice (via fix(compaction): use truncateUtf16Safe for post-compaction context text #102515), so this PR no longer touches that file.

Re-review

@clawsweeper re-review — addressed both P1 findings from the previous round:

1. Dropped post-compaction site — rebased onto current main 4be6fa7413; the import-churn conflict around post-compaction-context.ts is gone. The file is no longer touched by this PR.
2. Added after-fix real behavior proof — /tmp/proof-102630.mjs drives all four remaining sites with emoji-cross-boundary inputs. Site 1 (cap 57) and Site 3 (cap 79) demonstrate the bug-vs-fix difference (raw slice lone surrogate = true, helper lone surrogate = false); Sites 2 and 4 confirm length and surrogate safety on dense emoji input. Full output is in the PR body under "Real behavior proof (after-fix…)".

Pre-flight clean on the rebased head a8d0879089:

• node scripts/run-tsgo.mjs -p tsconfig.core.json --incremental — exit 0
• node scripts/test-projects.mjs on cli/plugins-list-format.test.ts, hooks/fire-and-forget.test.ts, auto-reply/fallback-state.test.ts, auto-reply/reply/conversation-label-generator.test.ts — 4 + 22 tests pass in 41.44s
• git diff --check — clean
• git diff origin/main --stat — 4 files changed, 8 insertions(+), 5 deletions(-)