adaptive-team-init

name	adaptive-team-init
description	Initialize the Adaptive Dev Team in your current project. Pulls files from the repo, discovers your codebase, and configures the team.
argument-hint	[optional: git ref/tag to pull, defaults to master]

Adaptive Team Init

Initializes the Adaptive Dev Team framework in your current project. Clones the repo, copies files, discovers your codebase, and configures the team — all in one flow.

Installed globally (~/.claude/skills/adaptive-team-init/) so it's available in any project.

Pipeline

Step 1: Check Dependencies

For each check, report the result and help fix if missing.

Required:

# Git installed
git --version

# Claude Code version (needs Agent Teams support)
claude --version
# If < v2.1.32: "Claude Code v2.1.32+ required. Run: claude update"

# Agent Teams enabled
cat ~/.claude/settings.json 2>/dev/null | grep -q "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS" \
  || cat .claude/settings.json 2>/dev/null | grep -q "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS"

If Agent Teams is not enabled, ask the user to enable it. Add to the appropriate settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Optional (report):

# Git repo
git rev-parse --is-inside-work-tree
# If not: "This directory is not a git repo. The team uses worktrees — initialize with git init?"

# Source files present
ls -A | grep -v "^\.git$" | head -1
# If empty: note the setup wizard will skip auto-detect; can re-run later once code exists.

Report a dependency summary before proceeding.

Step 2: Check for Existing Installation

Look for .claude/agents/adaptive-team-product-manager.md.

• If found: Ask — "Adaptive Dev Team is already installed. Overwrite with latest? (This won't touch your adaptive-team-context/ or adaptive-team-learned/ files.)"
• If declined: skip to the setup wizard (Step 6) to reconfigure.

Step 3: Clone to Temp

TEMP_DIR=$(mktemp -d)
git clone --depth 1 https://github.com/lwstory/adaptive-dev-team.git "$TEMP_DIR/adaptive-dev-team"

If a git ref/tag was provided, check out that ref.

Step 4: Copy Files

mkdir -p .claude/agents .claude/rules .claude/adaptive-team-context \
         .claude/adaptive-team-learned .claude/adaptive-team-state \
         .claude/adaptive-team-reviews

# Agent files (all adaptive-team-*.md)
cp "$TEMP_DIR/adaptive-dev-team/.claude/agents/adaptive-team-"*.md .claude/agents/

# Rules
cp "$TEMP_DIR/adaptive-dev-team/.claude/rules/adaptive-team-rules.md" .claude/rules/

# Skills (consult, implement, learning-moment — init is already installed globally)
for skill in adaptive-team-consult adaptive-team-implement adaptive-team-learning-moment; do
  mkdir -p ".claude/skills/$skill"
  cp "$TEMP_DIR/adaptive-dev-team/.claude/skills/$skill/SKILL.md" ".claude/skills/$skill/"
done

# Seed runtime directories from templates (only if missing)
# Note: runtime dirs (adaptive-team-learned/, adaptive-team-state/, adaptive-team-reviews/)
# are NOT committed to the framework repo. They hold per-project accumulated state. Templates
# in .claude/templates/ are the canonical starting content; this step materializes them locally.
mkdir -p .claude/adaptive-team-learned .claude/adaptive-team-state .claude/adaptive-team-reviews

# Lesson seed files (only if missing — never overwrite accumulated lessons)
for lesson in pm-lessons architect-lessons sdet-lessons llm-lessons database-lessons dev-lessons team-lessons curious-lessons; do
  [ ! -f ".claude/adaptive-team-learned/$lesson.md" ] && \
    cp "$TEMP_DIR/adaptive-dev-team/.claude/templates/.md.template" 



[ ! -f  ] && \
    .claude/adaptive-team-state/current-session.md
[ ! -f  ] && \
    .claude/adaptive-team-state/pending-pm-lessons.md


[ ! -f  ] && \
    .claude/adaptive-team-reviews/README.md


[ ! -f  ] && \
    .claude/


[ ! -f  ] && \
    


 -p docs
[ -f  ] && \
    docs/


[ -z  ] &&  .claude/adaptive-team-context/.gitkeep

Never overwrite adaptive-team-context/ or adaptive-team-learned/ contents.

Step 5: Clean Up Temp

rm -rf "$TEMP_DIR"

Step 6: Auto-Discover Project

Scan the project to detect tech stack and pre-populate defaults:

• Language/framework: read package.json, *.csproj, pyproject.toml, Cargo.toml, go.mod, etc.
• Project structure: list directories (2 levels), identify source vs test vs docs
• Test setup: find test dirs, config, frameworks, mock patterns
• Build system: Dockerfiles, CI/CD, build scripts
• Infrastructure: migrations, message queues, caches

Auto-detect defaults:

• package.json → npm test, Jest/Vitest
• *.csproj → dotnet test, xUnit/NUnit
• pyproject.toml → pytest
• Cargo.toml → cargo test
• go.mod → go test ./...

Step 7: Targeted Questions

Present auto-detected defaults. Only ask what can't be inferred:

1. Architecture style — Clean Architecture, MVC, microservices, monolith, serverless?
2. Code conventions — DI patterns, error handling, config management?
3. Test naming convention — (pre-fill from detected tests if possible)
4. Coverage targets — 80% overall? 90% core? No formal targets?
5. Build/verify command — (pre-fill, confirm or override)
6. Documentation locations — (pre-fill if docs/ detected)
7. Common quality issues — what problems recur?
8. Merge strategy — PRs or direct merge?

Step 8: Write Context Files

File	Content
tech-stack.md	Languages, frameworks, versions, infrastructure
architecture.md	Patterns, conventions, dependency rules
testing.md	Pyramid targets, naming, coverage, mock patterns, build command
review-checklist.md	Stack-specific review items for the architect

Step 9: Validate

• Each context file under 100 lines
• No duplication across files
• Content specific enough to be useful

Step 10: Confirm

Adaptive Dev Team initialized and configured.

Files installed:
  - 6 agent files (PM, architect, sdet, dev, llm-expert, database)
  - 1 rules file
  - 3 skills (consult, implement, learning-moment)
  - 7 learning file seeds
  - session state template, reviews README
  - project settings.json (if not already present)

Project context written:
  - tech-stack.md: [summary]
  - architecture.md: [summary]
  - testing.md: [summary]
  - review-checklist.md: [summary]

Ready to go. Use:
  /adaptive-team-consult <topic>    — design/approach discussion, no code
  /adaptive-team-implement <task>   — full implementation pipeline
  /adaptive-team-learning-moment    — capture a lesson from an issue

What Gets Copied

Copied	Not Copied
Agent files (adaptive-team-*.md)	README.md, CLAUDE.md, LICENSE
Rules (adaptive-team-rules.md)	.gitignore
Skills (consult, implement, learning-moment)	adaptive-team-context/ contents (generated fresh)
Learning file seeds (only if missing)	Project-specific learned content (preserved)
Session state template, reviews README (only if missing)
Project settings.json (only if missing)

Global Installation

~/.claude/skills/adaptive-team-init/SKILL.md

Install:

mkdir -p ~/.claude/skills/adaptive-team-init
curl -o ~/.claude/skills/adaptive-team-init/SKILL.md \
  https://raw.githubusercontent.com/lwstory/adaptive-dev-team/master/.claude/skills/adaptive-team-init/SKILL.md