Team Setup

team-setup

Overview

team-setup is an interactive skill that guides you through setting up the team AI directives. It presents four modes, explains each option, confirms your choice, and executes the setup.

It is invoked in two ways:

• User-invoked (/team-setup) — anytime, to configure or check a project.
• Model-invoked by team-boot — automatically at session start when a project has no .adlc/init-options.json configuration (self-install), so an unconfigured project wires itself without the user knowing the command.

The skill is non-destructive: it never overwrites existing files or directories. If the target path already contains a configured team AI directives, it detects this and offers the "Already configured" mode instead.

When to Use

• Starting a new team from scratch and need a neutral team AI directives scaffold to fill in later.
• Your team already has a directives repo on GitHub and you want to clone it locally.
• You have a local team AI directives directory already (e.g., from a previous project) and want to wire it up.
• You're unsure whether the team AI directives is already configured and want a quick check.
• When the project isn't yet wired to a team AI directives (no .adlc/init-options.json team_ai_directives field).
• Automatically via team-boot when it detects an unconfigured project at session start (self-install).

Decline Handling (when model-invoked by team-boot)

When team-boot invokes this skill because the project is unconfigured, the user may choose not to set up team AI directives right now. Handle decline explicitly to avoid a re-prompt loop:

• If the user declines at mode selection, do not run any mode. Exit cleanly and tell team-boot the user declined.
• Offer a persistent opt-out: "Don't ask again for this project?" On yes (build mode only), write .adlc/init-options.json with team_ai_directives: null:

  echo '{"team_ai_directives": null}' > ".adlc/init-options.json"

  This marker makes team-boot skip setup silently on every future prompt.
• In plan/read-only mode, a persistent opt-out cannot be written — the decline is session-scoped only; tell team-boot to defer.
• Never force a mode; the setup is user-consented at every step.

Core Process

Goal

Set up a team AI directives using one of four modes.

Security: Input Validation (all modes)

Before executing any mode, validate every user-supplied value (paths, URLs, team names). These values are interpolated into shell commands; unvalidated input is a command-injection vector.

• Paths ({DEST}, {ABSOLUTE_PATH}): reject if they contain any of `, $, ;, |, &, (, ), <, >, newline, or backslash. Resolve to an absolute path with realpath/Resolve-Path before use.
• Team name: must match ^[A-Za-z0-9 ._-]+$. Reject anything else.
• Clone URL (Mode 1): must start with https://. Reject file://, ssh://, and any non-https scheme unless the user explicitly confirms the risk. Cloning runs no code from the repo, but the cloned content is read by agents later — only clone repositories you trust.

If any value fails validation, report which value and why, and re-ask. Never interpolate a user value into a Python/eval source string — pass it through the environment (see Mode 2).

Fast path: if .adlc/init-options.json already contains a valid team_ai_directives path, skip to Mode 4 (Already Configured) — do not re-clone, re-point, or re-scaffold.

Modes at a Glance

Mode 1: Clone from GitHub

Full walkthrough in references/mode-clone.md.

Mode 2: Point to Existing Local Path

Full walkthrough in references/mode-local.md.

Mode 3: Scaffold New Empty team AI directives

Full walkthrough in references/mode-scaffold.md.

Mode 4: Already Configured

Full walkthrough in references/mode-configured.md.

Mode Selection Flow

1. Explore: Present the user with four options:

   How would you like to set up team-ai-directives?
   
   1) Clone from GitHub — Clone an existing repository
   2) Point to existing local path — Use a team AI directives you already have
   3) Scaffold new empty team AI directives — Create a fresh neutral team AI directives
   4) Already configured — Check existing configuration

2. Present: For the chosen mode, explain what will happen and show details.

3. Confirm: Ask the user to confirm before executing.

4. Write/Execute: Perform the setup for the chosen mode.

Post-Setup Configuration

After any mode completes successfully, update the project configuration:

1. Write team_ai_directives to .adlc/init-options.json
2. Verify the team AI directives is accessible by running a quick health check:
   • {TEAM_AI_DIRECTIVES}/context_modules/constitution.md exists
   • {TEAM_AI_DIRECTIVES}/.skills.json exists and is valid JSON
3. Check the .gitignore convention (ADR-401 R7 allowlist — see the Gitignore Convention Check step below).
4. Inject the project-level AGENTS.md directive — full command and managed-section contract in references/post-setup-agents.md.
5. Install MCP config — merge details in references/post-setup-mcp.md.

Gitignore Convention Check (ADR-401 R7)

Verify .gitignore follows the ADLC allowlist — a fresh team setup must never leave a wholesale .adlc/ ignore in place (it would hide the tracked .adlc/ artifacts the team model relies on):

1. If .gitignore contains a bare .adlc/ rule, replace it (and any .adlc/* + !.adlc/... lines that conflict) with the R7 allowlist: ignore .adlc/*, .adlc/evals/results/, .adlc/memory/*, .adlc/team-learn-report.md, .adlc/team-levelup-report.md, and graphify-out/, while re-including !.adlc/init-options.json, !.adlc/workspace.yml, !.adlc/drafts/, !.adlc/evals/, !.adlc/memory/, !.adlc/memory/evals/, and !.adlc/memory/evals/holdout.json. The canonical rule list lives in the workspace skill's scripts/bash/paths.sh (GITIGNORE_RULES_ALLOWLIST, mirrored in scripts/powershell/paths.ps1).
2. Also verify the agent-install surface rules are ignored (.agents/, .opencode/, .claude/, .cursor/, .codex/, .gemini/, .qwen/, .devin/, .tabnine/, skills-lock.json, .skills.json, .mcp.json, .events.json, .pytest_cache/, .ruff_cache/).
3. Never remove unrelated (non-ADLC) rules; only add missing allowlist rules.
4. Tracked legacy files keep working — the allowlist gates only untracked files, so no migration is required before the switch.