ChaseFlorell/agent-team

A generic, full-stack team of Claude Code role agents and the skills they
preload. Every file here names nothing specific to one project, so the team
works in any repository; a project's own rules live in that project's
companion skills, which name the generic skill they extend and win where they
differ.
The team
| Agent | Name | Role |
|---|---|---|
| spec-author | Jennifer | business analyst — scenarios and what is out of scope |
| test-writer | Kevin | QA engineer — a claim into a failing step definition |
| backend | Brad | back-end engineer — server side, scripts, build, test, and check CI |
| ux | Tiffany | UX designer — interaction flow, accessibility semantics, and localized copy on every platform; builds nothing |
| client | Megan | client engineer — web, desktop, and mobile clients, built from the UX hand-off and approved boards |
| designer | Brittany | UI designer — visual layout, boards, states, and design-system values before build |
| infrastructure | Dave | cloud and DevOps engineer |
| database-administrator | Jane | PostgreSQL schema, EF Core mapping, migrations |
| spec-reviewer | Jessica | code reviewer — a diff against its claims and decisions |
| critic | Karen | design reviewer — a plan before it is built |
| adversary | Kyle | security engineer — attacks a change before it merges |
| auditor | Ashley | compliance auditor — the whole repository at rest |
| babysitter | Kristy | build babysitter — watches pull requests and builds, classifies each failure, routes it |
| ai-author | Emily | maintains agent instructions |
Each agent's frontmatter declares its model, effort, tools, and the skills it
preloads, all under skills/. An agent with no tools key inherits every tool
the session offers, connectors included; the designer relies on that to reach
a design tool, so a tools list added to it must keep those.
The babysitter guards its commands with a frontmatter hook, so spawn it as a
plain subagent, never as an agent-team teammate; teammates get no hooks.
Team skills
| Skill | Holds | Use when |
|---|---|---|
| agent-persona | how a named role agent speaks: voice in session, plain findings | acting as a named role agent |
| build-avalonia-ui | Avalonia layer of build-native-ui: resources, control themes, pseudo-class states, automation properties | implementing or changing an Avalonia screen, component, or theme |
| build-native-ui | stack-agnostic native build to the hand-off: tokens, states, platform accessibility APIs, component and UI tests | implementing or changing a native screen, view model, or component on any stack |
| build-web-ui | the client's web build, exactly to the hand-off: accessible, localized, responsive markup, component tests, screenshots | implementing or changing a web screen or component |
| clarify-requirements | resolving a genuinely material requirement ambiguity | two readings would build different systems |
| coding-conventions | orient first, stop on a gap, keep the design direct, put a rule where it runs | any code, test, doc, or diagram change |
| deliver-change | issue, worktree, pull request, docs, and CI workflow | creating or editing issues, worktrees, PRs, or checks |
| design-avalonia-ui | Avalonia layer of design-native-ui: tokens, Light and Dark variants, control states, density, drawn versus templated, boards, builder hand-off | designing or changing an Avalonia screen, component, or theme |
| design-cloud-infrastructure | infrastructure as code: a diagram, least privilege, cost, rollback, the plan workflow, telemetry, dashboards, and alerts | designing or changing infrastructure, secrets, backups, deployment, telemetry, dashboards, or alerts |