Enhance prompt injection detection by adding word boundaries for know…

[prashant4654]

This pull request makes a targeted improvement to the regular expression used for detecting known jailbreak personas in the agentflow/utils/validators.py file. The change adds word boundaries to the regex pattern, which helps avoid false positives when matching persona names like "APOPHIS", "STAN", and "DUDE".

Regex improvements for persona detection:

• Updated the regex pattern for known jailbreak personas to include word boundaries, ensuring only standalone words are matched and reducing false positives. (agentflow/utils/validators.py)

[Copilot]

Pull request overview

This PR refines the prompt-injection validator’s “known jailbreak persona” detection regex in agentflow/utils/validators.py by adding word boundaries, reducing false positives from substring matches (e.g., matching STAN inside STANford).

Changes:

• Updated the known-persona regex to use \b word boundaries and a non-capturing group for APOPHIS|STAN|DUDE.

[codecov]

Codecov Report

❌ Patch coverage is 64.28571% with 10 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
agentflow/utils/converter.py	59.09%	8 Missing and 1 partial ⚠️
agentflow/graph/agent_internal/providers.py	50.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Iamsdt]

Move that formatting logic inside convert message function