harness: never cut a surrogate pair when truncating tool-result previews - #150
Open
Alexsun1one wants to merge 1 commit into
Open
harness: never cut a surrogate pair when truncating tool-result previews#150Alexsun1one wants to merge 1 commit into
Alexsun1one wants to merge 1 commit into
Conversation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
previewTexttruncates tool-result previews with a plain.slice(0, TOOL_RESULT_PREVIEW_CHARS), which cuts by UTF-16 code units. When an emoji (or any astral character) straddles the 4000-char boundary, the slice keeps only the high surrogate. The lone surrogate serializes as an unpaired\ud8xxescape, and a strict JSON parser consuming the serialized event downstream rejects the whole message — one emoji at the boundary fails the entire turn.This is a production repro, not a hypothetical: my agent signs every reply with 🚒🕵️, so its conversation history is full of surrogate pairs. A scheduled task that read that history back through a tool hit the boundary and the turn died with
invalid TypeScript harness protocol message ... unexpected end of hex escape.Fix
After slicing, drop a trailing lone high surrogate before appending the
...[truncated]marker.previewTextis exported so the boundary behavior can be unit-tested.Tests
Four cases in
tools.test.ts: short text unchanged, plain truncation, an emoji straddling the boundary (fails on the old code), and a pair that fits entirely inside the boundary staying intact.Follow-up thought
This fixes the known producer, but as akrentsel noted on Discord, the executor could also tolerate a corrupted/unparseable protocol line more gracefully than failing the turn (skip + log, or surface a structured error event). Happy to take a stab at that separately if there's interest — it touches the Rust side, so I kept it out of this PR.