Team Onboarding Template

Welcome to [Team Name]

How We Use Claude

Based on Alex’s usage over the last 30 days (2,228 sessions across 320+ repo projects, one repo dominant at ~47%):

Work Type Breakdown (tool-mix proxy):
  Build Feature     ████████████████████  38%
  Analyze Data      ████████████░░░░░░░░  23%
  Plan Design       ███████████░░░░░░░░░  20%
  Debug Fix         ████████░░░░░░░░░░░░  12%
  Improve Quality   ████░░░░░░░░░░░░░░░░   7%

Top Skills & Commands:
  /mcp                         ████████████████████  10x/month
  /effort                      ████████████████████  10x/month
  /bad                         ████████████████████  10x/month
  /extra-usage                 ██████████████████░░   9x/month
  /ultraplan                   ████████░░░░░░░░░░░░   4x/month
  /oh-my-claudecode:team       ██████░░░░░░░░░░░░░░   3x/month
  /oh-my-claudecode:autopilot  ██████░░░░░░░░░░░░░░   3x/month
  /loop                        ██████░░░░░░░░░░░░░░   3x/month
  /model                       ██████░░░░░░░░░░░░░░   3x/month
  /compact                     ████░░░░░░░░░░░░░░░░   2x/month
  /gravy                       ████░░░░░░░░░░░░░░░░   2x/month

Top MCP Servers:
  computer-use       ████████████████████  ~427 calls
  chrome-devtools    ████████████████░░░░  ~345 calls
  oh-my-claudecode   ████████░░░░░░░░░░░░  ~164 calls
  historian          █░░░░░░░░░░░░░░░░░░░  ~11 calls
  context7           ░░░░░░░░░░░░░░░░░░░░  occasional

Top tool calls (past 30 days): Bash 24.6k · Read 11.8k · Edit 4.9k · Grep 3.5k · Glob 1.4k · Write 1.3k · Agent 903 · Task* 884 · ToolSearch 364 · Skill 161 · WebFetch 157.

Your Setup Checklist

Codebases

• [primary-repo] — [repo url] # primary day-to-day work
• [secondary-repo] — [repo url] # orchestration / tooling
• [integration-repo] — [repo url] # external-service integration
• [other team repo] — [repo url]

MCP Servers to Activate

• computer-use — Desktop automation (screenshots, clicks, typing) for native macOS/Windows apps. Install the computer-use MCP package; approve app access per workflow.
• chrome-devtools — Browser automation through the Chrome DevTools Protocol. Used for scraping, digesting tabs, and driving web apps. Install the Chrome DevTools MCP server.
• oh-my-claudecode (OMC) — Multi-agent orchestration layer with skills like autopilot, ralph, team, ultrawork. Run omc setup or /oh-my-claudecode:omc-setup.
• context7 — On-demand library/SDK docs lookup. Configure via MCP settings.

Skills to Know About

• /mcp (~10x/mo) — Manage MCP connections mid-session.
• /effort (~10x/mo) — Bump reasoning effort when stuck on genuinely hard debugging or architecture problems. Default stays low to conserve tokens.
• /bad (~10x/mo) — BMAD autonomous development pipeline: parallel story implementation, each story isolated in its own worktree. Trigger for “run BAD” / “automate the sprint”.
• /extra-usage (~9x/mo) — Check your quota before kicking off heavy work.
• /ultraplan (~4x/mo) — Consensus planning across Planner / Architect / Critic before execution. Use on high-risk changes.
• /oh-my-claudecode:team (~3x/mo) — N coordinated Claude agents on a shared task list; stage-aware routing.
• /oh-my-claudecode:autopilot (~3x/mo) — Full autonomous execution from idea to working code. Use for broad “build me X” requests.
• /loop (~3x/mo) — Run a prompt or slash command on a recurring interval. Good for polling builds, scheduling self-paced tasks, or babysitting PRs.
• /compact (~2x/mo) — Compress conversation history mid-session before context rot kicks in.
• /check-env — Pre-flight environment check (ports, Docker, .env, git status) before starting dev work.
• /schedule — Remote scheduled agents (cron triggers on claude.ai cloud) for daily kickoffs, dep audits, PR babysitting.
• /codex:rescue — Hand a stuck or genuinely hard task to Codex for a second implementation pass.

Team Tips

• Verify before claiming done. Run the build, the test, or the actual workflow — not just read the diff. Paste command + last 20 lines of output + exit code.
• Never push, PR, or deploy without explicit permission. Auto mode is not a license to ship; same for destructive git ops.
• Check the environment first. Run /check-env before starting dev — ports, Docker, .env, git state. Most “it doesn’t work” moments are environmental.
• Run npx tsc --noEmit after multi-file TS changes. Consider a hook to run this automatically on Edit/Write.
• Use skills, not ad-hoc prompts, for repeating workflows. /oh-my-claudecode:autopilot for broad builds, /loop for recurring checks, /schedule for cloud cron, /codex:rescue when stuck.
• Stay on low reasoning effort by default. Use /effort to escalate only for genuinely hard problems — tokens are finite.
• Record fixes in tasks/lessons.md. Format: - **[date] Problem**: ... → **Fix**: .... Promote recurring patterns to CLAUDE.md weekly.
• Read the Claude playbook. Live at https://ao92265.github.io/claude-code-playbook/, source at https://github.com/ao92265/claude-code-playbook. Covers skills, hooks, MCPs, workflows.

Get Started

Work through these in order:

1. Read the playbook. Skim https://ao92265.github.io/claude-code-playbook/ end-to-end before writing your own prompts.
2. Environment warm-up. Clone [primary repo], run /check-env, then ask Claude to summarize the repo architecture from AGENTS.md (or run /oh-my-claudecode:deepinit if it doesn’t exist yet).
3. Scheduled-trigger smoke test. Ask Claude to create a daily kickoff trigger via /schedule and walk you through the first digest. Teaches the cloud-agent model without touching local code.
4. Lessons-driven absorption. Read tasks/lessons.md in the primary repo (if it exists) and ask Claude to explain the top 3 recurring problems and how to avoid them.