---
description: AI / agent / LLM workflow standards (AgentX)
globs: **/*agent*,**/*llm*,**/*model*,**/*workflow*,**/agents/**,**/*ai*
alwaysApply: false
---

# AI / Agents / LLM (AgentX)

Canonical: `.github/instructions/ai.instructions.md` -- read for full standards.

- Pin model versions; define evaluation strategy and guardrails up front.
- Treat prompts as versioned artifacts; add tracing/observability hooks.
- Validate and sanitize model inputs/outputs; guard against prompt injection.