grounded-ai-dev

Cross-cutting AI/LLM guardrails and evaluation checklist for Grounded AI engagements. Use on any build with an LLM, agent, RAG pipeline, chatbot, copilot, or generative feature — during Specify (AI requirements) and Implement (guardrails and evals). Trigger when working on prompts, model selection, hallucination control, evals, agent actions, or AI safety.

Master controller for Active AI Advisors software development. Use this skill the moment ANY request involves building, developing, coding, prototyping, or shipping software — an app, a web app, a tool, an automation, an integration, a feature, an MVP, a script, an agent, a chatbot, a dashboard, or any AI/LLM-powered product. Trigger on phrases like "build," "create an app," "develop," "write a tool," "automate," "make a prototype," "stand up a service," "add a feature," or any request that would produce running code for a client or for AAA. This skill enforces the six-phase Grounded AI process and the full-lifecycle quality gates, and routes to the phase and review skills. Do not write production code without it.

Phase 2 of Grounded AI — CHALLENGE. Use after a Requirements Document exists and before any tests or code, to stress-test the specification for gaps, ambiguities, and edge cases. Trigger when grounded-specify completes or when the user asks to review or pressure-test requirements. Produces a Specification Review Report and an updated Requirements Document (v2).

Phase 6 of Grounded AI — DOCUMENT. Use at delivery to produce the documentation package that lets the client understand, operate, and evolve the system independently. Trigger when verification is complete or when the user asks to document, hand over, or package a delivery. Produces the Grounded AI Delivery Package.

Phase 4 of Grounded AI — IMPLEMENT. Use after tests are approved (Gate B) to build code that satisfies the specification and passes the test suite. Trigger when grounded-test-first completes and tests are approved, or when the user asks to build, code, or implement against an approved spec and tests. Produces the delivered codebase with review notes.