fix: prevent panics when truncating non-ASCII strings in agent log helpers#8024
Draft
cgarrett-square wants to merge 5 commits intoblock:mainfrom
Draft
fix: prevent panics when truncating non-ASCII strings in agent log helpers#8024cgarrett-square wants to merge 5 commits intoblock:mainfrom
cgarrett-square wants to merge 5 commits intoblock:mainfrom
Conversation
…response merges After a tool executes and the subsequent LLM call fails with a provider error (network, credits, context, etc.), the error message was yielded to the frontend but never added to messages_to_add. The session would end with user(tool_response) as the last message. On the next user turn, fix_conversation's merge_consecutive_messages would combine that tool_response user message with the new user message into a single mixed-content message, which is invalid for OpenAI-format providers. This caused an irrecoverable loop: every new message the user sent triggered the same provider rejection. Fix 1 (agent.rs): add the error message to messages_to_add before breaking in all provider error branches so the session always ends with an assistant message after a failed LLM call. Fix 2 (conversation/mod.rs): do not merge consecutive user messages when either contains ToolResponse content — tool results must stay in their own message so providers can correlate them with the corresponding tool requests. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…lpers Byte-index slicing (`&s[..N]`) panics if the Nth byte falls in the middle of a multi-byte UTF-8 character (e.g. emoji). Two sites in agent.rs were affected: - `derive_call_reason`: truncating user message preview at byte 100 - `format_messages_for_log`: truncating tool response body at byte 2000 Replace both with `char_indices().nth(N)` to find a safe char boundary before slicing, eliminating the runtime panic. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Covers the two panic sites fixed in the previous commit: - derive_call_reason: truncation of user message preview at char 100 - format_messages_for_log: truncation of tool response body at char 2000 Each site gets three cases: ASCII truncation, short (no truncation), and a multi-byte input (emoji / CJK) that would have panicked before the fix. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…austive error - Replace char_indices byte-boundary slicing with chars().take(N).collect() to satisfy the clippy::string_slice lint (no byte-index slicing at all) - Construct CallToolResult via serde_json deserialization in test to avoid the #[non_exhaustive] restriction on struct literals outside rmcp crate Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
DOsinga
reviewed
Mar 20, 2026
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
2 participants
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.
Summary
Two log helper functions in
agent.rsused byte-index slicing (&s[..N]) to truncate strings, which panics at runtime if the cutoff byte falls in the middle of a multi-byte UTF-8 character (e.g. emoji, accented characters, CJK text).This caused
tokio-runtime-workerthread panics that killed the agent mid-session whenever a user message or tool response contained non-ASCII content near the truncation boundary.Affected sites:
derive_call_reason— truncated user message preview at byte 100format_messages_for_log— truncated tool response body at byte 2000Fix
Replace byte-index slicing with
char_indices().nth(N)to find a safe UTF-8 character boundary before slicing. This is the standard Rust idiom for character-count-based truncation.Test plan