extractUserRequest matches the trigger phrase mid-token, extracting the wrong text as the user's request

[ManoharPaturi]

extractUserRequest matches the trigger phrase mid-token, extracting the wrong text as the user's request

Type: bug (trigger/extraction mismatch)
Severity: low-medium
Area: src/utils/extract-user-request.ts
Effort: trivial

Summary

The trigger matcher and the request extractor disagree about what counts as
an occurrence of the trigger phrase.

checkContainsTrigger (src/github/validation/trigger.ts) only accepts the
phrase at a word boundary — preceded by start-of-text or whitespace and
followed by whitespace/punctuation/end:

const regex = new RegExp(
  `(^|\\s)${escapeRegExp(triggerPhrase)}([\\s.,!?;:]|$)`,
  "i",
);

But extractUserRequest (src/utils/extract-user-request.ts) takes the
first raw substring match:

const triggerIndex = commentBody
  .toLowerCase()
  .indexOf(triggerPhrase.toLowerCase());

An occurrence embedded inside a token — which the matcher correctly refuses
to treat as a trigger — therefore wins the extraction.

Failure scenario

Comment body:

Email security@claude.dev ASAP. @claude please review the auth module

• checkContainsTrigger: security@claude.dev is rejected (preceded by
  y), the second occurrence matches → the action runs.
• extractUserRequest: indexOf finds @claude inside security@claude.dev
  first → the "user's request" becomes
  .dev ASAP. @claude please review the auth module.

The prompt then leads with wrong text; at best the model recovers, at worst
it acts on garbage. The same applies to any earlier mention of the phrase in
a URL, code span, or username-like token (@claude-bob, cc @claude/bot).

Suggested fix

Scan occurrences left to right and honor only the first one whose preceding
and following characters satisfy the same boundary semantics as the trigger
matcher, keeping the file's string-operations approach (no RegExp built from
the phrase). Happy to open a PR.