Best possible solution:

Land this narrow helper substitution and regression test if CI remains green; keep broader UTF-16 hardening as separate site-specific fixes rather than expanding this branch.

Do we have a high-confidence way to reproduce the issue?

Yes. Source inspection of current main shows raw slice at the truncation boundary can split a surrogate pair, and the added test constructs an agent id boundary that exercises the failure path; I did not run tests in this read-only review.

Is this the best way to solve the issue?

Yes. Updating the existing limitText helper to use the shared UTF-16-safe truncation facade is the narrowest maintainable fix because both final agent capsules and workspace context snippets already flow through that helper.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against c5d42593be6d.

Label changes:

• add proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes copied terminal output from the current head showing the affected runtime path returns hasLoneSurrogate:false, plus the focused Vitest command/output for the regression test.
• add rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🐚 platinum hermit and patch quality is 🦞 diamond lobster.
• add status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (terminal): The PR body includes copied terminal output from the current head showing the affected runtime path returns hasLoneSurrogate:false, plus the focused Vitest command/output for the regression test.

Label justifications:

• P3: The PR fixes a low-blast-radius voice-call prompt-formatting edge case without changing config, auth, migrations, provider routing, or delivery semantics.
• rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🐚 platinum hermit and patch quality is 🦞 diamond lobster.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (terminal): The PR body includes copied terminal output from the current head showing the affected runtime path returns hasLoneSurrogate:false, plus the focused Vitest command/output for the regression test.
• proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes copied terminal output from the current head showing the affected runtime path returns hasLoneSurrogate:false, plus the focused Vitest command/output for the regression test.

PR surface:

Source +1, Tests +25. Total +26 across 2 files.

What I checked:

• Current main still has the bug-prone boundary: On origin/main, limitText still truncates with text.slice(0, Math.max(0, maxChars - 32)), and buildRealtimeVoiceInstructions feeds both workspace context and the final agent capsule through that helper. (extensions/voice-call/src/realtime-agent-context.ts:25, c5d42593be6d)
• PR uses the established helper: The PR changes the truncation boundary to truncateUtf16Safe(...) from openclaw/plugin-sdk/text-utility-runtime, keeping the marker/cap behavior otherwise unchanged. (extensions/voice-call/src/realtime-agent-context.ts:26, 7b5f6d8f9328)
• Helper contract is suitable for bundled plugins: The plugin SDK facade exports truncateUtf16Safe, whose implementation floors the limit and delegates to sliceUtf16Safe so a trailing high surrogate is dropped instead of returned alone. (packages/normalization-core/src/utf16-slice.ts:44, c5d42593be6d)
• Runtime entry point checked: Voice-call runtime builds realtime instructions once before constructing the realtime handler, so this helper is on the touched session setup path rather than an unused formatter. (extensions/voice-call/src/runtime.ts:301, c5d42593be6d)
• Sibling truncation pattern checked: Adjacent voice-call realtime consult truncation already uses truncateUtf16Safe before adding the same kind of truncation marker, which supports this as the local consistency fix rather than a new policy surface. (extensions/voice-call/src/webhook/realtime-handler.ts:196, c5d42593be6d)
• Focused proof and regression test: The PR body includes copied terminal output from head 7b5f6d8f93 showing hasLoneSurrogate:false, and the diff adds a regression test that places an emoji at the truncation boundary for buildRealtimeVoiceInstructions. (extensions/voice-call/src/realtime-agent-context.test.ts:97, 7b5f6d8f9328)

Likely related people:

• Solvely-Colin: The realtime agent-context module and its limitText helper appear to date to the merge for fix ios qr scanner lifecycle #101235, authored by @Solvely-Colin. (role: introduced behavior; confidence: high; commits: 632efa2d7343; files: extensions/voice-call/src/realtime-agent-context.ts, extensions/voice-call/src/runtime.ts)
• vincentkoc: @vincentkoc authored the merged UTF-16 bounded-output hardening at fix(text): keep bounded outputs UTF-16 safe #101355, which updated adjacent voice-call truncation sites with the same helper pattern. (role: recent adjacent contributor; confidence: medium; commits: 84e5327720b4; files: extensions/voice-call/src/media-stream.ts, extensions/voice-call/src/webhook.ts, src/talk/consult-question.ts)
• Alix-007: The current PR author also appears as a co-author on the merged sibling UTF-16 hardening commit for fix(text): keep bounded outputs UTF-16 safe #101355, so they have relevant recent context beyond this proposal. (role: recent adjacent contributor; confidence: medium; commits: 84e5327720b4; files: extensions/voice-call/src/media-stream.ts, extensions/voice-call/src/webhook.ts)

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

• reviewed 2026-07-07T03:53:00.544Z sha edf8c84 :: needs real behavior proof before merge. :: none