Free Claude Code Pack
A Claude Code Agent Team
That Ships While You Sleep
One prompt, five agents: a planner, a researcher, a builder, a reviewer, and a shipper — each in its own context. The drop-in setup that stops one giant chat from losing the plot and burning tokens. Free, copy-paste ready.
What Are Claude Code Subagents?
Subagents are specialized Claude Code agents that live in .claude/agents/, each defined by a markdown file with its own instructions, tool permissions, and model. The key detail: each subagent runs in its own context window. When the main thread delegates a task, the subagent does the work in a clean context and returns only the result — so a 40,000-token codebase sweep never pollutes your main session.
This pack gives you five working subagents wired into a team: an orchestrator that plans and delegates, a researcher that maps the code, an implementer that writes the change, a reviewer that tries to break it, and a shipper that runs the tests and verifies.
Why Does One Big Claude Code Chat Fail on Real Work?
A single context is a generalist with amnesia. It re-reads the same files, accumulates irrelevant history, loses earlier decisions halfway through, and ships a change that looks right and isn't. The fix isn't a cleverer prompt — it's stopping doing it all in one chat. Split the work across specialists with clean desks.
One giant chat
- Context bloats with every turn
- Re-reads files it already saw
- Forgets decisions made 30 turns ago
- No second pass — ships plausible-but-wrong
- Token cost compounds fast
An agent team
- Each agent gets a clean context window
- Researcher hands forward only what matters
- Implementer works from a tight brief
- Reviewer catches the wrong change before it ships
- Main thread stays small and cheap
The core setup
How Do I Set Up a Claude Code Agent Team?
Drop a .claude/agents/ folder and a CLAUDE.md into any repo. Each agent is one markdown file with frontmatter — name, description, the tools it's allowed to use, and the model — followed by its system prompt. The CLAUDE.md sets the delegation rule: the orchestrator passes forward the artifact, not a summary, and runs research → implement → review → ship in order.
The free pack below is all of it, ready to paste. Drop it in, fill the deploy line, and say "use the orchestrator to build X."
orchestrator
Plans, decomposes, delegates in order. Never edits code itself. The entry point for any multi-step task.
researcher
Read-only scout. Maps the codebase and cites file:line so the implementer never re-explores.
implementer
Writes the change in the smallest diff that matches the repo, and verifies what it touched.
reviewer
Adversarial review against the plan. Blocking vs non-blocking findings. Read-only.
shipper
The final gate: runs tests, build, and deploy/verify. Reports failures faithfully.
CLAUDE.md
Wires the team together, sets the standards, and holds the delegation rule that saves tokens.
What's the Difference Between Subagents and the Main Agent?
The main agent is the thread you talk to. Subagents are workers it dispatches with the Task tool. The main agent keeps the conversation and the plan; subagents get a single job, their own context, and a restricted toolset — then return a result. You stay in control: the orchestrator reports back what shipped, what was skipped, and any risk.
The practical win is least privilege. The researcher and reviewer have no edit tools at all, so they can't accidentally change code. Only the implementer writes. That one decision removes a whole class of mistakes.
Do Subagents Actually Save Tokens?
Yes, when you delegate correctly. The expensive part of agentic coding is context that gets re-sent on every turn. A subagent does its heavy reading in a separate window and returns a short artifact, so that reading never re-enters the main thread's context on subsequent turns. The orchestrator passing forward the distilled result — not the raw sweep — is what keeps the main session small.
It also improves quality, which is the real point. The mandatory review step catches the plausible-but-wrong change a single pass would have shipped. Cheaper and more correct.
Video walkthrough
Watch the agent team build a feature end to end.
The full walkthrough — research → implement → review → ship, with the token savings and the review step catching a bug — drops on YouTube. Subscribe to catch it.
Subscribe on YouTubeFree download
Get the Agent Team Starter Pack
All five agents — orchestrator, researcher, implementer, reviewer, shipper — plus the CLAUDE.md that wires them together and a 60-second install guide. Copy-paste ready, drops into any repo.
Want me to build it for you?
AI Workflow Audit — $750
The pack gives you the team. The audit covers your entire Claude Code setup — CLAUDE.md review and rewrite, MCP server audit, your own custom agent team designed around your actual work, and a prioritized action list delivered as a 75-minute Google Meet walkthrough with copy-paste configs.
Most setups I audit are doing everything in one bloated context, burning tokens and shipping changes nobody reviewed. The pack fixes the obvious. The audit builds the system around how you actually work.
75-minute Google Meet. Copy-paste deliverables. No upsells.
Guarantee: if the audit does not surface at least 5 actionable improvements to your Claude setup, full refund. Same day.
FAQ
Common questions
What is a Claude Code subagent?+
How many subagents should I use?+
Do subagents save tokens?+
Where do the agent files go?+
What model should each agent use?+
Is this free?+
More free resources
