fix: keep stage and subagent text on UTF-8 character boundaries

[BigUncleUs]

Summary

Fixes two byte-level text truncations in the agent stage / notification path that can split a multi-byte UTF-8 character and emit an invalid byte sequence to downstream consumers. In non-English (CJK / emoji) deployments this surfaces as replacement characters (�) and, depending on the consumer's UTF-8 handling, can corrupt or break the message (e.g. a strict WebSocket text-frame validator rejecting the frame).

Both issues are present on current master (4f9e2c0).

Changes

• claw_core: claw_core_publish_stage_tool_calls() truncated tool-call arguments_json with "%.40s", cutting at a raw byte offset. The file already has a utf8_prefix_len() helper (used for the reasoning snippet just above), so the fix reuses it and the cut now lands on a character boundary.
• claw_manager: claw_agent_mgr_completion_observer() splices a [truncated] suffix into the <subagent_completed> buffer at a fixed byte offset on overflow, which can leave an orphaned partial UTF-8 character in the kept prefix. The splice offset is now backed off to a character boundary before the suffix is written.

Implementation

• claw_core: compute prefix_len = utf8_prefix_len(args, 40) and print with "%.*s"; the ellipsis is shown when strlen(args) > prefix_len. This mirrors the existing reasoning-snippet path, so no new helper is introduced.
• claw_manager: before writing the suffix, decrement the splice offset while the byte at that position is a UTF-8 continuation byte ((byte & 0xC0) == 0x80), so the retained prefix always ends on a character boundary. The strlcpy size is widened to the real remaining buffer.
• No new public API; both changes are local to existing functions.

Files

File	Change
components/claw_modules/claw_core/src/claw_core_events.c	Reuse utf8_prefix_len() for tool-call args truncation instead of "%.40s"
components/claw_modules/claw_manager/src/claw_agent_mgr.c	Back the [truncated] splice offset off to a UTF-8 boundary

Test plan

• Symptom reproduced end-to-end in a downstream integration: CJK tool args and long sub-agent final_text produced � / invalid UTF-8 in the stage and <subagent_completed> text; gone after the fix.
• Built in a clean ESP-IDF environment (pure logic change, no toolchain locally; relying on CI for the component build)
• Unit test added (none added; behavior verified manually)

[CLAassistant]

All committers have signed the CLA.

[Copilot]

Pull request overview

This PR fixes two truncation paths that previously operated on raw byte offsets and could split multi-byte UTF-8 sequences, resulting in invalid UTF-8 being emitted in stage/notification text.

Changes:

• claw_core: truncate tool-call arguments_json using utf8_prefix_len() + "%.*s" so the cut lands on a UTF-8 character boundary.
• claw_manager: when inserting the "[truncated]" suffix into <subagent_completed>, back the splice position off continuation bytes to avoid cutting into a multi-byte UTF-8 sequence.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
components/claw_modules/claw_manager/src/claw_agent_mgr.c	Adjusts truncation suffix splice point to preserve UTF-8 boundaries in <subagent_completed> notifications.
components/claw_modules/claw_core/src/claw_core_events.c	Reuses existing utf8_prefix_len() to safely truncate tool-call argument previews on character boundaries.

---

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

[laride]

sha=61ea5660818f34cabb2269d7d1685884e43ef64b

[laride]

Thank you for your PR! We are currently going through our internal review process.