Best possible solution:

Keep the UTF-16 helper approach, but make chunkString preserve all input for every positive limit it accepts or explicitly reject/clamp unsupported tiny limits, with a limit-1 regression test.

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

Yes, source-reproducible. Current main uses raw code-unit slicing at both reported boundaries, and the current PR head still exposes a focused chunkString limit-1 data-loss case.

Is this the best way to solve the issue?

No, not fully as submitted. Reusing the existing UTF-16 helpers is the right owner-boundary fix, but the empty-safe-slice branch should preserve the full code point or reject/clamp tiny limits instead of skipping input.

Full review comments:

• [P3] Preserve tiny-limit surrogate chunks — src/agents/bash-tools.shared.ts:145-146
  The current head still advances past an empty safe slice without emitting anything. For chunkString("🚀x", 1), sliceUtf16Safe(input, 0, 1) returns "", this branch increments i by 2, and the joined chunks become "x", so the exported helper drops user output for a positive limit. Please emit the full code point, clamp/reject unsupported tiny limits, or otherwise preserve input.
  Confidence: 0.86

Overall correctness: patch is correct
Overall confidence: 0.84

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against eae3fc1a79dd.

Label justifications:

• P2: This is a focused agent-runtime correctness fix for output and resume-message text handling with limited blast radius.
• rating: 🦐 gold shrimp: Overall readiness is 🦐 gold shrimp; proof is 🦞 diamond lobster and patch quality is 🦐 gold shrimp.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (terminal): The PR body includes terminal output from a standalone production-path proof plus focused tests showing the before/after UTF-16 boundary behavior.
• proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes terminal output from a standalone production-path proof plus focused tests showing the before/after UTF-16 boundary behavior.

PR surface:

Source +11, Tests +70. Total +81 across 4 files.

Acceptance criteria:

• [P1] node scripts/run-vitest.mjs src/agents/bash-tools.shared.test.ts src/agents/task-resume-utf16-safe.test.ts src/agents/subagent-orphan-recovery.test.ts.

What I checked:

• Current main bash chunking: Current main chunks sanitized bash stdout/stderr through chunkString, and chunkString still uses raw input.slice(i, i + limit), which can split a surrogate pair at the fixed transport boundary. (src/agents/bash-tools.shared.ts:140, eae3fc1a79dd)
• Current main resume truncation: Current main builds the orphaned-subagent resume task text with task.slice(0, maxTaskLen), so a task ending with a split surrogate pair at the 2000-code-unit boundary can produce a replacement character. (src/agents/subagent-orphan-recovery.ts:75, eae3fc1a79dd)
• UTF-16 helper contract: sliceUtf16Safe backs out a slice endpoint that would expose only part of a surrogate pair, and existing tests assert sliceUtf16Safe(input, 0, 1) returns an empty string for a leading emoji pair. (packages/normalization-core/src/utf16-slice.ts:33, eae3fc1a79dd)
• PR head remaining defect: At PR head, an empty safe slice with a positive tiny limit advances i without pushing the skipped code point, so chunkString("🚀x", 1).join("") would lose the leading emoji. (src/agents/bash-tools.shared.ts:145, 99b58c454ea7)
• PR test gap: The added chunkString tests cover the default 8192-code-unit boundary and a limit-2 case, but not the limit-1 empty-safe-slice case that still drops data. (src/agents/bash-tools.shared.test.ts:97, 99b58c454ea7)
• Contributor proof: The PR body includes focused Vitest output and terminal production-path proof for the default chunk and resume-message UTF-16 boundary behavior; the live PR also carries the proof: sufficient label. (99b58c454ea7)

Likely related people:

• vincentkoc: Blame and PR metadata show the current affected helper files and normalization helper locations were localized by PR refactor(line): localize internal declarations #102037, authored and merged by this account. (role: recent area contributor; confidence: high; commits: 63192752325b; files: src/agents/bash-tools.shared.ts, src/agents/subagent-orphan-recovery.ts, packages/normalization-core/src/utf16-slice.ts)
• steipete: History shows this account merged the original UTF-16 truncation fix and recently refactored adjacent bash exec helper code in PR refactor: consolidate markdown code fences, error coercion, and byte-identical helper pairs #99932. (role: adjacent owner; confidence: medium; commits: 63f5fa47deb6, 49cc59b1e88e; files: src/utils.ts, src/agents/bash-tools.exec-runtime.ts)
• amittell: Recent merged work on preserving subagent task text during restart redispatch touched the same orphan-recovery flow adjacent to buildResumeMessage. (role: recent area contributor; confidence: medium; commits: 7b5d86e89a5f; files: src/agents/subagent-orphan-recovery.ts, src/agents/subagent-registry-run-manager.ts)
• jalehman: Recent transcript-reader refactor work touched the subagent recovery flow around last-message extraction and resume context. (role: adjacent contributor; confidence: medium; commits: d216f7c876dd; files: src/agents/subagent-orphan-recovery.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 2026-07-08T07:24:24.458Z sha 52d7059 :: needs changes before merge. :: [P3] Preserve tiny-limit surrogate chunks
• reviewed 2026-07-09T00:46:05.896Z sha 99b58c4 :: needs changes before merge. :: [P3] Preserve tiny-limit surrogate chunks