Support prompt usage

[steven10a]

Fixes reported bug from rcatoen:

TypeError: GuardrailAgent.__new__() missing 1 required positional argument: 'instructions' when using prompt instead of instructions

Fix:

• Typing for instruction was too strict required it to be a string
• Functionality was already supported as we are simply a wrapper around Agent SDK

Also added support for users adding their own input and output guardrails via the Agent SDK.

• These were silently suppressed and only guardrails via the config were run
• We now merge the lists and run all of them

Deleted an old NSFW.md file that is not being used

[Copilot]

Pull Request Overview

This PR enhances the GuardrailAgent class to support dynamic instructions and user-provided guardrails, while also removing outdated NSFW benchmark documentation.

• Made the instructions parameter optional and support callable functions for dynamic generation
• Added support for user-provided input_guardrails and output_guardrails that merge with config-based guardrails
• Removed deprecated NSFW benchmark documentation file

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/guardrails/agents.py	Enhanced GuardrailAgent.__new__() to accept optional/callable instructions and merge user-provided guardrails with config-based guardrails
docs/benchmarking/nsfw.md	Removed entire file containing outdated NSFW benchmark results and documentation

---

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

[gabor-openai]

LGTM thank you!