AI Setup Flow

Status: Coming soon

Agent Teams includes three bundled AI agents that cover the entire project lifecycle, from initial setup to ongoing team evolution. The dashboard surfaces each one contextually — only when it is relevant to your current state.

Invocation syntax — the bundled agents work in both GitHub Copilot Chat and Claude Code, but the syntax differs by platform:

Platform	Syntax	Example
GitHub Copilot Chat	@agent-name (chat participant)	@project-configurator
Claude Code	/agent-name (slash command)	/project-configurator

The rest of this document shows both forms where invocation is mentioned. Use whichever matches your active AI tool.

---

The Three-Agent Flow

# GitHub Copilot Chat
@project-configurator  →  @agent-designer  →  @consultant

# Claude Code
/project-configurator  →  /agent-designer  →  /consultant

      (configure)           (create agents)    (advise & evolve)

Each agent is independent and can be used on its own, but together they form a progressive onboarding path that takes you from zero to a fully configured, AI-ready workspace.

The flow is intentionally incremental: start with a minimal set of focused agents, then let @consultant identify gaps and guide you through adding more. This avoids front-loading a complex team structure that consumes unnecessary tokens before you know what you actually need.

---

Step 1 — Project Configurator

When: no project.profile.yml exists yet Access: dashboard home → "Auto-configure with AI" button, or invoke directly — @project-configurator (Copilot) / /project-configurator (Claude Code)

@project-configurator performs an exhaustive, technology-agnostic analysis of your repository to auto-generate a complete project.profile.yml and a set of atomized context packs with real, codebase-derived content.

What it detects

The agent does not assume any technology. It reads the directory tree inductively and looks for any build manifest or project descriptor it finds — including but not limited to:

Ecosystem	Detected files
JavaScript / TypeScript	package.json, pnpm-workspace.yaml, turbo.json, nx.json
Rust	Cargo.toml, Cargo.lock
.NET / C#	*.csproj, *.sln, *.fsproj
Android / Kotlin	build.gradle, build.gradle.kts, settings.gradle, app.json
Go	go.mod, go.sum
Python	pyproject.toml, requirements.txt, Pipfile, setup.py
Swift / iOS	Package.swift, Podfile, *.xcodeproj, *.xcworkspace
Flutter / Dart	pubspec.yaml
Java / Kotlin (JVM)	pom.xml, build.gradle, build.sbt
Elixir	mix.exs
C / C++	CMakeLists.txt, Makefile
Infra	Dockerfile, docker-compose.yml, *.tf, Pulumi.yaml, cdk.json
CI/CD	.github/workflows/*.yml, .gitlab-ci.yml, Jenkinsfile

For monorepos with multiple sub-projects using different stacks, each workspace is analysed independently before the global profile is synthesised.

What it generates

After you confirm the proposal, the agent writes:

1. .agent-teams/context-packs/{name}.md — one pack per detected functional domain (e.g. android-app, backend-api, infra, conventions), plus a universal architecture pack. Every pack contains real content derived from the analysis: module responsibilities, data flow descriptions, architectural patterns observed, and code conventions — not boilerplate.

2. .agent-teams/project.profile.yml — all fields populated from evidence in the repository:

   • technologies — only techs with direct file evidence
   • commands — verbatim invocations from manifests (./gradlew assembleDebug, cargo test --workspace, etc.)
   • paths — mapped from the actual directory structure
   • sync_targets — inferred from which target directories already exist (.github/agents/ → github_copilot, .claude/agents/ → claude_code, etc.)

The agent always waits for your explicit confirmation before writing any file. If a project.profile.yml already exists, it shows a diff and asks before overwriting.

---

Step 2 — Agent Designer

When: profile configured, you want to create or edit an agent Access: Team Agents card → "Design with AI" button (primary), or @agent-designer (Copilot) / /agent-designer (Claude Code)

The "Design with AI" button is the primary action in the Team Agents card header and in the empty-state when no agents exist yet. The previous "Create manually" flow remains available as the secondary option.

@agent-designer reads your workspace context — existing agents, installed skills, and the project profile generated in step 1 — and produces a focused, validated AgentSpec. Start with one or two agents covering your core workflows; you can always add more later guided by @consultant.

See Agent Designer for the full workflow.

---

Step 3 — Consultant

When: profile + at least one agent exist Access: dashboard home → "Consult your team" card, or @consultant (Copilot) / /consultant (Claude Code)

@consultant is a read-only advisory agent. It analyses the current state of your team against the project profile and provides structured recommendations. It never modifies any file — execution always defers to @agent-designer.

What it analyses

• Coverage gaps — functional domains in the project profile with no dedicated agent
• Responsibility overlaps — pairs of agents sharing the same primary responsibility
• Topology issues — broken handoff references, missing escalation targets, circular delegation chains
• Context pack assignments — packs in .agent-teams/context-packs/ not referenced by any agent

What it recommends

• New agent proposals — concrete id, role, domain, intents, and scope.topics for each gap, grounded in the actual tech stack
• MCP integrations — external services where an MCP server would let an agent act rather than just advise (e.g. @modelcontextprotocol/server-github for a GitHub-facing agent)
• Skill assignments — skills from skills.registry.yml that should be added to specific existing agents, with when conditions
• Next steps — every response closes with explicit pointers: "To add data-engineer → open @agent-designer"

This advisory loop is how the team grows incrementally: add agents one at a time as @consultant identifies real gaps, rather than trying to design the full team upfront.

---

Dashboard Surfacing

The dashboard home page shows the right agent CTA at each stage:

State	Card shown	Primary action
No profile	Configure Your Project	Auto-configure with AI → @project-configurator (Copilot) / /project-configurator (Claude Code)
No profile	Configure Your Project	Configure manually → Profile Editor
Profile ✓	Team Agents header & empty state	Design with AI → @agent-designer (Copilot) / /agent-designer (Claude Code)
Profile ✓	Team Agents header & empty state	Create manually → Agent Wizard
Profile ✓, agents > 0	Consultant	Consult your team → @consultant (Copilot) / /consultant (Claude Code)

---

Using All Three Together

A typical first-time setup with a new repository:

1. Install the extension → dashboard opens with the "Configure Your Project" card
2. Click "Auto-configure with AI" → @project-configurator (Copilot) or /project-configurator (Claude Code) scans your repo, proposes the profile and context packs → confirm → files written
3. Dashboard detects the new profile → Team Agents card shows "Design with AI" → click it → @agent-designer / /agent-designer reads your profile and existing agents, produces a focused AgentSpec → save → repeat for each initial agent you need
4. Sync once → agents deployed to your target platforms
5. As the project grows, open "Consult your team" → @consultant / /consultant identifies coverage gaps → follow its "Next steps" pointers to open @agent-designer / /agent-designer for each new agent

All three agents work on any project type or language — the entire flow is technology-agnostic.