what-context-needed

Overview

A meta-prompting skill for GitHub Copilot that asks the AI to identify what files and context it needs before answering a question. Instead of guessing or hallucinating answers, Copilot first lists the files it requires, explains why each is relevant, and identifies uncertainties — resulting in more accurate and grounded responses.

Key Features

• Pre-answer context discovery: Before answering any question, Copilot outputs a structured list of files it needs to examine, categorized into "Must See" (required for accuracy), "Should See" (helpful for completeness), and "Already Have" (from conversation context).
• Transparent reasoning: Each requested file includes an explanation of why it is relevant to the question, making the AI's reasoning process visible and auditable.
• Uncertainty identification: Explicitly calls out what the AI is unsure about without seeing the actual code, helping users understand potential gaps in the response before it is generated.
• Two-pass interaction pattern: Users first invoke this skill with their question, provide the requested files, then ask the question again — producing significantly more accurate answers through informed context.

Who is this for?

This skill is ideal for developers working on large or unfamiliar codebases who want to reduce AI hallucination and improve answer quality. It is especially valuable when asking complex architectural questions, debugging across multiple files, or when Copilot's initial answers seem to miss important project-specific context. The two-pass approach trades a small amount of extra interaction time for substantially better accuracy.