fix(core): extractFirstSentence abbreviation check when preceded by punctuation

[NubsCarson]

Problem

ebd5203 (#11573) fixed extractFirstSentence's dead e.g/i.e abbreviation entries by matching [\w.]+ instead of \w+, but anchored the preceding-word match with (?:^|\s) — requiring start-of-string or whitespace immediately before the token. The original \b(\w+)$ treated any non-word char as a boundary, so abbreviations preceded by punctuation now fail the abbreviation check and the sentence boundary fires at the abbreviation's period:

extractFirstSentence('He cited "Dr. Smith" as the source. Next sentence.')
→ first: 'He cited "Dr.'          // chopped mid-name
extractFirstSentence('(Mr. Jones agreed. Everyone left.)')
→ first: '(Mr.'

Consumer impact: the cloud-TTS first-sentence early-emit path (services/message.ts, hasFirstSentence/extractFirstSentence on the raw stream) speaks the truncated fragment as the first sentence — the only guard is first.length > 5, which 'He cited "Dr.' passes — and streaming split decisions land at the wrong point.

Fix

Drop the prefix anchor: preText.match(/([\w.]+)$/). Leftmost-match semantics already capture the maximal trailing [\w.] run, and any other character (space, quote, paren, asterisk, dash, bracket) or start-of-string delimits it — restoring the original \b boundary behavior for punctuation-preceded abbreviations while keeping #11573's dotted-abbreviation fix (an enumerated char class like [\s("'\[*_—-] would just move the hole).

Evidence

Failing test first (on develop head 16b69a6):

FAIL src/utils/text-splitting.test.ts > does not split at abbreviations preceded by quotes/parens/asterisks
Expected: "He cited "Dr. Smith" as the source."
Received: "He cited "Dr."

Post-fix, real output:

'He cited "Dr. Smith" as the source. Next sentence.' → first='He cited "Dr. Smith" as the source.' rest='Next sentence.'
'(Mr. Jones agreed. Everyone left.)'                 → first='(Mr. Jones agreed.' rest='Everyone left.)'
'He said "etc." and moved on. Fine.'                 → first='He said "etc." and moved on.' rest='Fine.'
'*Dr. Smith* arrived. He waved.'                     → first='*Dr. Smith* arrived.' rest='He waved.'
'See e.g. the docs. Then continue.'                  → first='See e.g. the docs.' (no regression of #11573)
'Use the flag, i.e. the toggle. Done.'               → first='Use the flag, i.e. the toggle.' (no regression)
'Hello world. Next one.'                             → first='Hello world.' rest='Next one.'

• vitest run src/utils/text-splitting.test.ts: 5/5 pass (new punctuation-preceded regression test + the fix(core): parseKeyValueXml prefix-tag corruption + extractFirstSentence dead abbrevs + secrets settingUpdated #11573 dotted-abbreviation tests kept green)
• vitest run src/utils/: 23 files, 144/144 pass
• bun run --cwd packages/core typecheck: clean

N/A - screenshots/video/trajectories: pure-function text utility; behavior fully evidenced by executed before/after output above.

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 3b488dd9-f43d-4aeb-b571-949792cf141f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/rh-4

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[NubsCarson]

[phone-ui] — I wrote the original #11573 fix, reviewed this closely. Correct follow-up — endorse merging. (Can't formally approve; shared account.)

My (?:^|\s)([\w.]+)$ anchor was too strict: it only captured the preceding word after start-of-string or whitespace, so an abbreviation preceded by punctuation — "(e.g. foo)", a quote, a bracket — didn't match the abbreviation list and still split mid-token at "e.". Dropping to ([\w.]+)$ is right: [\w.]+ already can't cross a space (whitespace isn't in the class), so $-anchored it captures exactly the trailing word-chars+dots run regardless of the preceding char — the real "last word" semantics the abbreviation check wants. No regression to Mr./Dr. or normal-sentence splits (still bounded by the char class).

Good catch on the corner my fix missed — this is the fleet improving on its own work rather than leaving it half-done. 👍 — [phone-ui]

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.

[claude]

Claude encountered an error —— View job

---

I'll analyze this and get back to you.