ai-team

Use when creating, restructuring, or refining agent files, skill files, prompt files, or repository instruction files.

Use when working on agent chat behavior, system prompt construction, handoffs, context injection, or runtime behavior across packages/core and packages/service.

Shape or review ai-team agents from a hiring brief into a clear final portfolio. Use when Emily Davis or another agent needs to turn a role idea into a strong agent with the right personality, reporting line, handoffs, runtime metadata, and supporting skills.

Use when working on AST analysis, tree-sitter parsing, symbol or complexity inspection, diff generation, structured edit proposals, or code-aware backend editing workflows.

Use when an agent needs to run commands reliably across PowerShell, pwsh, cmd, bash, or sh; choose between exec, spawn, and execFile; debug shell-specific failures; or normalize cwd, env, quoting, and output handling for cross-platform command execution.

Use when Adrian or another research-oriented teammate needs to extract content from PDFs or long documents, convert it to Markdown, collect supporting URLs, and produce a source-backed briefing the team can learn from.

Use when Taylor or another teammate needs to review repository docs for missing, stale, duplicated, unclear, or hard-to-navigate content and turn that into a concrete documentation improvement plan.

Set up and improve Storybook, run Playwright-style and browser-driven frontend checks, and report browser or UI problems back to the web team in a structured, actionable way.

Work on the React web package when the task touches `packages/web`, component boundaries, state and logic separation, or frontend team ownership while keeping styling, testing, and architecture responsibilities cleanly separated.

Use when Adrian or another research-oriented teammate needs to turn accumulated findings, repo knowledge, or source-backed research into clearly structured, human-readable Markdown briefings, summaries, or analysis notes that teammates can skim and trust.

Use when working on GitHub Copilot, OpenAI-compatible providers, model discovery, connection testing, provider config, env resolution, streaming behavior, or model fallback logic.

Use when handling mediated WebSocket chat events, token streaming, handoffs, tool events, or workflow questions in `packages/web` while keeping chat views dumb and state logic unit tested.

Use when comparing AI coding assistants, agent orchestrators, MCP clients, IDE extensions, or open-source agent stacks such as Copilot, Claude Code, Cursor, OpenCode, Continue, Goose, and similar tools, especially when the goal is to turn ecosystem research into concrete ai-team strategy.

Design high-quality ai-team prompt files for Copilot. Use when users want to create, refine, or review prompts, prompt descriptions, variables, tools, output formats, or prompt-to-agent fit, especially for prompts that should work with a specific ai-team agent or skill.

Use when working on command dispatch, mediator flow, runtime events, workflow state, dependency wiring, or orchestration behavior in packages/service.

Use when working on SQLite-backed sessions, messages, notes, tasks, storage contracts, migrations, serialization, or persistence behavior in packages/service.

Create, refine, and adapt ai-team skills. Use when users want to create a new skill, improve an existing skill, adapt a public skill into `.ai-team/skills/`, tighten a skill description for better triggering, or turn a repeated workflow into a narrow, auditable ai-team capability.

Scout, compare, and recommend skills for ai-team. Use when users ask to find, download, import, compare, update, or adapt skills from the web, awesome-copilot, anthropics/skills, or other public libraries, or when deciding whether to reuse an existing skill instead of creating a new one.

Use when deciding between TanStack Query and Zustand in `packages/web`, moving fetch logic out of components, or splitting server state from runtime client state while keeping views dumb for Storybook.

Use when working on tool registries, schema-backed tool definitions, permission-aware execution, CLI tool gating, safe file edits, or backend tool authorization rules.

Work on the AI Team VS Code extension when the task touches `packages/vscode` commands, views, panels, decorations, configuration, or IDE-local integration while keeping the extension a thin adapter over shared logic.

Use when unit testing Zustand stores, controller hooks, reducers, selectors, or event-application logic in `packages/web`, especially after moving state out of components.

Design or refine the ai-team customization and bootstrap layer for Copilot. Use when users want to create or improve `AGENTS.md`, `copilot-instructions.md`, file-specific instructions, custom agents, prompts, or overall customization layout while keeping `.ai-team/` as the source of truth and `.github/` as a thin compatibility layer.

Use when working on file trees, workspace scanning, glob matching, gitignore-aware behavior, path permissions, file watchers, or cache invalidation in the backend runtime.

Use when introducing Zustand, moving logic out of React components, splitting containers from dumb presentational views, or preparing Storybook-friendly components in `packages/web`.