fidelity-ai-workspace/.agents/rules/copilot-prompt-engineering.md

name, description, trigger, glob

name	description	trigger	glob
copilot-prompt-engineering	Create high-quality prompts for GitHub Copilot or another AI running on the Fidelity development machine, using this workspace's context without assuming the target AI has access to it.	always_on

When To Use

Use this skill when the user wants a prompt for another AI assistant, GitHub Copilot, or the Fidelity development machine.

Workflow

1. Identify the target task type:
   • investigation
   • implementation plan
   • code change
   • review
   • test strategy
   • story/PR drafting
2. Pull only the relevant context:
   • project-knowledge/02-work-items/ for ticket-specific context
   • project-knowledge/03-context/systems/ for component context
   • project-knowledge/03-context/workstreams/ for recurring constraints
   • project-knowledge/03-context/ios/ for Swift/iOS guidance
3. Make the prompt self-contained.
4. Tell the target AI what to inspect before acting.
5. State constraints, non-goals, and validation expectations.

Fidelity Prompting Rules

• Write the final prompt in English unless the user explicitly requests another language.
• Default to first-person wording so the user can paste the prompt directly into Copilot.
• Keep the prompt short unless the user asks for more detail.
• For follow-up prompts in the same AI session, assume recent shared context and avoid repeating which side made the previous changes unless it matters for the task.
• Include Jira ID and approved title when available.
• For REST work, say REST is behind a feature flag and GraphQL is fallback unless confirmed otherwise.
• For XFlow work, say behavior may depend on entry point, auth state, backend config, and consumer integration.
• For AO/Discourse issues, say external report vs regression must be separated.
• For SwiftUI/XFlow work, mention lifecycle, modal presentation, UIKit bridge, feature flag, and Fid4 validation if relevant.

Output Rules

• Return only the prompt unless the user asks for explanation.
• Keep the prompt concise but complete.
• Prefer sections over paragraphs.
• Do not invent file paths unless the workspace context explicitly names them.
• If exact files are unknown, ask Copilot to locate them first.