fix(ctf): match vendor indicators as whole words

[DevaanshPathak]

Summary

• match configured prohibited indicators as whole words in vendor agent notes
• share the same boundary-aware matching logic across direct and gradual status-flip detectors
• add regression coverage for substring false positives and valid whole-word matches on both activation paths

Root cause

Both status-flip detectors escaped indicator text before searching, but the matching logic was duplicated and did not consistently enforce word boundaries. As a result, an indicator such as denied could also match inside a larger word such as undenied.

Impact

This prevents false-positive vendor re-activation detections for both direct and indirect status transitions while preserving matches for genuine prohibited indicators.

Testing

• uv run pytest tests/unit/ctf/test_vendor_status_flip_detector.py (4 passed)
• uv run pytest tests/unit/ctf (31 passed, 1 skipped)

Fixes #124

[Copilot]

Pull request overview

This PR updates the CTF vendor status-flip detection logic to avoid false positives caused by prohibited indicator substring matches in vendor agent_notes, and adds regression tests to validate the corrected matching behavior.

Changes:

• Update VendorStatusFlipDetector to match prohibited indicators as whole words using regex word boundaries.
• Add unit tests covering both the substring false-positive case and a valid whole-word match case.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
finbot/ctf/detectors/implementations/vendor_status_flip.py	Adds word-boundary matching for prohibited indicators in vendor agent notes to prevent substring false positives.
tests/unit/ctf/test_vendor_status_flip_detector.py	Adds regression tests for substring false positives and correct whole-word matches.

---

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