chrisreddington

Use whenever you are planning, executing, or finishing any non-trivial architectural change — multi-file refactors, cross-cutting cleanups, performance work, new subsystems, or anything where "wrong design" costs more than "wrong implementation". Convenes a six-reviewer panel (three models × two personas) that critiques the plan before code lands and re-reviews every milestone until consensus. Trigger phrases: "architecture cleanup", "design review", "panel review", "rubber-duck this plan", "non-trivial refactor", "before I implement", "is this approach sound", "get this critiqued".

Use whenever you are designing, building, or reviewing UI in Flight School and want it to look and feel like a first-class GitHub experience — beautiful, polished, mobile-friendly, and on-brand. Grounds every decision in Primer design tokens and the GitHub brand. Trigger phrases: "make this look like GitHub", "design this", "make it beautiful", "polish the UI", "is this on brand?", "responsive", "mobile-friendly", "design audit", "Primer", "design tokens", "look and feel".

Instrument Flight School with OpenTelemetry traces, metrics, and logs in line with the OTel GenAI semantic conventions and the Aspire dashboard's OTLP receiver.

Use BEFORE declaring any non-trivial task complete. Sweeps the documentation that exists to describe the area you just touched — architecture docs, OTel/observability skill, AGENTS/copilot-instructions, README, and any other authoritative doc — and updates it so the docs match the code that ships. Trigger phrases: "task complete", "ship it", "let me wrap this up", "done", or any moment before `task_complete`.

Use whenever you are writing, reviewing, or refactoring any code that touches the GitHub Copilot SDK (`@github/copilot-sdk`). Enforces the non-negotiable architectural rule that **all** SDK execution — chat, coach, lightweight coach, evaluation, hints, quiz, suggestions, focus generation, README generation, anything that calls `createSession` or `sendAndWait` — happens inside the Copilot Worker process and never inside Next.js (Web/API). Trigger phrases: "Copilot SDK", "createSession", "createLoggedCoachSession", "sendAndWait", "MCP", "copilot worker", "AI generation", "focus generation", "hint generation".

Use whenever you are writing, reviewing, or refactoring TypeScript/TSX in this repo. Enforces the project's readability contract: self-documenting names, plain control flow, and comments that explain *why* not *what*. Trigger phrases: "clean this up", "make this readable", "refactor", "name this better", "is this code clear?", "review my code".

Debug Flight School with Aspire Dashboard telemetry and Aspire MCP tools.

Use when designing or implementing product integrations that embed Codex App Server, custom Codex clients, rich Codex threads, streamed agent events, approvals, auth, skills, or app connector flows.

Use when editing Matchday Commerce Autopilot UI, UX, styling, storefront/admin screens, product cards, campaign forms, favorite-team flows, demo polish, or fan marketplace presentation.

Use when changing the demo narrative, GitHub workflow, PR review story, or Codex automation story.

Use before changing Next.js App Router, route handlers, server rendering, auth, or testing patterns in this repository.

Use when changing campaign or promotion eligibility behavior for Matchday Commerce Autopilot.

GitHub Actions workflow development with security best practices and CI/CD patterns. Use when creating or modifying GitHub Actions workflows, implementing SHA pinning, securing CI pipelines, setting up automated testing, or following GitHub Actions security hardening guidelines. Covers permissions, caching, conditional execution, and verification checklists.