fix(guard): use UTF-8 safe string truncation in output preview logging

[lpcox]

Summary

Fixes #3711 — the WASM guard panics on multi-byte UTF-8 characters in tool response previews, poisoning the entire session.

Root cause

label_response in lib.rs truncates serialized JSON for debug logging using byte-index slicing (&output_json[..500]). When byte 500 falls in the middle of a multi-byte UTF-8 code point (CJK = 3 bytes, emoji = 4 bytes), Rust panics with "byte index is not a char boundary". Since this runs inside the WASM guest, the panic becomes a trap that permanently poisons the guard instance.

Changes

New helper — safe_preview(s, max_bytes) -> &str:

• Uses str::floor_char_boundary() (stable since Rust 1.80) to find the nearest valid character boundary at or before the byte limit
• Returns the full string when shorter than the limit

Three call sites fixed in label_response:

Location	Before	After
Path-specific output preview (~L808)	&output_json[..500] — panics	safe_preview(&output_json, 500)
General output preview (~L939)	&output_json[..500] — panics	safe_preview(&output_json, 500)
Input preview (~L752)	from_utf8(&bytes[..500]) — silent drop	safe_preview(full_str, 500) — always logs

8 unit tests covering:

• ASCII strings (under, at, and over the 500-byte limit)
• CJK characters (3-byte UTF-8) crossing the boundary
• Emoji (4-byte UTF-8) crossing the boundary
• Accented characters (2-byte UTF-8) at exact boundary
• Mixed ASCII + CJK content simulating real JSON payloads
• Empty string edge case

Evidence

Discovered in moeru-ai/airi PR triage workflow (run #24311673575) — a PR with a Chinese body caused the guard to crash, and all subsequent MCP calls failed with "WASM guard is unavailable after a previous trap".

Verification

make agent-finished passes — all unit + integration tests green.

[Copilot]

Pull request overview

This PR fixes a session-killing panic in the GitHub WASM guard caused by truncating &str log previews at a raw byte offset that may split multi-byte UTF-8 characters. It introduces a UTF-8-safe preview helper and updates label_response logging to use it, plus adds focused unit tests to prevent regressions.

Changes:

• Added safe_preview(s, max_bytes) and a shared PREVIEW_MAX_BYTES constant for UTF-8-safe log truncation.
• Replaced unsafe string slicing in label_response output previews with safe_preview.
• Added unit tests covering ASCII and multi-byte UTF-8 boundary cases (CJK, emoji, accented characters, mixed content, empty string).

Show a summary per file

File	Description
guards/github-guard/rust-guard/src/lib.rs	Adds UTF-8-safe truncation helper, updates preview logging call sites to avoid panics, and introduces unit tests for boundary conditions.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 1/1 changed files
• Comments generated: 3