- Use code-point length instead of UTF-16 length in estimateStringChars()
so that CJK Extension B+ surrogate pairs (U+20000+) are counted as 1
character, not 2 (fixes ~25% overestimate for rare characters).
- Change long-line split step from maxChars to chunking.tokens so that
CJK lines are sliced into token-budget-sized segments instead of
char-budget-sized segments that produce ~4x oversized chunks.
- Add tests for both fixes: surrogate-pair handling and long CJK line
splitting.
Addresses review feedback from Greptile and Codex bots.
The QMD memory system uses a fixed 4:1 chars-to-tokens ratio for chunk
sizing, which severely underestimates CJK (Chinese/Japanese/Korean) text
where each character is roughly 1 token. This causes oversized chunks for
CJK users, degrading vector search quality and wasting context window space.
Changes:
- Add shared src/utils/cjk-chars.ts module with CJK-aware character
counting (estimateStringChars) and token estimation helpers
- Update chunkMarkdown() in src/memory/internal.ts to use weighted
character lengths for chunk boundary decisions and overlap calculation
- Replace hardcoded estimateTokensFromChars in the context report
command with the shared utility
- Add 13 unit tests for the CJK estimation module and 5 new tests for
CJK-aware memory chunking behavior
Backward compatible: pure ASCII/Latin text behavior is unchanged.
Closes#39965
Related: #40216
Webhook channels (LINE, Zalo, Nextcloud Talk, BlueBubbles) are
incorrectly flagged as stale-socket during quiet periods because
snapshot.mode is always undefined, making the mode !== "webhook"
guard in evaluateChannelHealth dead code.
Add mode: "webhook" to each webhook plugin's describeAccount and
propagate described.mode in getRuntimeSnapshot.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Wrap the embedded agent stream to catch 'Unhandled stop reason: ...'
errors from the provider adapter and convert them into structured
assistant error messages instead of crashing the agent run.
Covers all unknown stop reasons so future provider additions don't
crash the runner. The wrapper becomes a harmless no-op once the
upstream dependency handles them natively.
Fixes#43607
