Add Guardrails Setup Agent Skill

[joshreini1]

Summary

TruLens supports blocking guardrails (evaluate before returning a response) for both LangChain and LlamaIndex, but the existing skills don't cover guardrail configuration. Users who want to use TruLens feedback functions as runtime safety checks have no guided workflow.

What

Create skills/guardrails/SKILL.md that walks users through:

• Choosing which feedback functions to use as guardrails (safety metrics like harmfulness/toxicity are obvious, but context_relevance as a hallucination gate is powerful too)
• Configuring threshold and action (block, warn, fallback response)
• Framework-specific setup for LangChain (WithFeedbackFilterDocuments) and LlamaIndex guardrail patterns
• Testing guardrails with adversarial inputs
• Monitoring guardrail trigger rates in the dashboard

Reference

• Guardrail code: src/apps/langchain/trulens/apps/langchain/guardrails.py and src/apps/llamaindex/trulens/apps/llamaindex/guardrails.py
• Quickstart: examples/quickstart/blocking_guardrails.ipynb
• Skills spec: structured Markdown (SKILL.md), version 0.1.0, lives in skills/ directory

Difficulty

Easy-Medium