The Ascent/Learn
The Ascent · Learn

Level 4 · Orchestrate — Claude Code

Direct a team

A team of agents works in parallel under rules you wrote. You set the destination, the order and the policy — not the keystrokes.

The lineYou can’t watch every keystroke, and a written rule isn’t enforced. You govern through the brief, the rules file, the order of play and the merge gate.
13 screens ≈ 15 min 10 floors to climb Full reference
Scroll, swipe or press ↓ · T light / darkBegin 01 / 13
The AscentL4 · Orchestrate · What it is

01 What this level is

Govern the team, don’t type the work

You brief one lead agent. It splits the job across specialists that run at the same time.

  • You write the law

    A rules file every agent reads before it starts.

  • Out of the steps, on the loop

    You don’t watch the keystrokes; rules and checks do.

  • You hold the gate

    Nothing merges until you’ve read what changed.

Brief Build the Q1 board dashboard from the raw data — under the rules file.
GM · lead agent — splits the job, assigns rooms, assembles running in parallel
Dataclean + aggregate the CSVlive aggregates ✓idle
Webbuild the charts — holds for Data4 charts wired ✓idle
Reviewertrace every figure; flag gaps to you1 figure with no source → flagged to youidle
Result dashboard branch → your merge gate

A rule with no check is a wish — you read the diff before anything merges.

Press Run: independent work runs at once; Web holds until Data lands.

The Ascent · Learn · OrchestrateNext 02 / 13
The AscentL4 · Orchestrate · Rooms

02 Folders as departments

Every folder is a room

At this level the project is the building and its subfolders are the rooms. Claude Code reads a room’s CLAUDE.md on top of the project’s when it works there.

  • 1

    The house rules

    The root AGENTS.md, imported by CLAUDE.md — every agent, every session.

  • 2

    A room’s brief

    data/CLAUDE.md: only what’s local — read when Claude works in data/.

  • 3

    Skills

    Packaged procedures any room can load.

  • 4

    Subagents

    Specialists the lead hands work to.

ExplorerEXAMPLE
board-dashboard (folder)
  • .claude (folder)
    • skills (folder)
      • brand-styles (folder) — a skill
    • agents (folder)
      • fact-checker.md — a subagent
  • design (folder)
  • data (folder)
    • CLAUDE.md — data rules only
  • web (folder)
  • AGENTS.md — The house rules
  • CLAUDE.md — one line: @AGENTS.md
The Ascent · Learn · OrchestrateNext 03 / 13
The AscentL4 · Orchestrate · How it loads

03 The rules file

How the rules load

  • Every session starts fresh

    The file re-briefs Claude, every time.

  • No more retyping

    Rules you’d otherwise repeat are always there.

  • One brief for everyone

    Same rules for every person and agent.

  • It improves with use

    Claude repeats a mistake? Add a line.

Guidance, not a lock. Keep it short and specific. Hard limits belong in checks and settings.
LOAD ORDER · BROAD → SPECIFIC
  1. Org policy
    IT-managed · optional
  2. Your personal CLAUDE.md
    ~/.claude
  3. Project AGENTS.md
    via CLAUDE.md
  4. Subfolder file
    only when Claude works there
  5. Claude’s own notes (auto memory)
    this machine only
  6. Your prompt
    e.g. “Refresh the board dashboard…”
All of the above adds up to: Claude’s context for this sessionAdded together, not overwritten.
The Ascent · Learn · OrchestrateNext 04 / 13
The AscentL4 · Orchestrate · File, skill, agent

04 Which one?

House rules, a skill or an agent?

All three start as plain Markdown files. What differs is when Claude uses them, and what they hold.

AGENTS.md

The house rules

Loaded

Every session in the folder

Holds

Who we are, house style, folders, guardrails

Lives in

The top of the project folder

Example
- Never invent figures.
- Outputs go in outputs/.

Skill

The playbook for one task

Loaded

Only when the task comes up, or you type /name

Holds

Steps, templates, examples, scripts

Lives in

.claude/skills/

Example

/weekly-summary drafts the Friday update

Subagent

A specialist you hand work to

Loaded

When Claude delegates a job to it

Holds

Its own instructions and tools; reports back

Lives in

.claude/agents/

Example

A fact-checker that traces every figure

illustrative
Rule of thumb
Applies to everythingAGENTS.md
A procedure for one jobSkill
A big job to hand offSubagent
The Ascent · Learn · OrchestrateNext 05 / 13
The AscentL4 · Orchestrate · Skills

05 Skills

Repeatable work, packaged

A skill is a folder: a SKILL.md with the steps, plus any templates, examples or scripts.

  • Always visible

    Its name and short “use when…” description.

  • Loaded only when needed

    Claude matches your request, then reads the full steps.

  • Or call it directly

    Type /weekly-summary to run it by name.

SKILL.mdEXAMPLE
EXPLORER
weekly-summary
SKILL.md
template.md
examples
good-summary.md
scripts
check_totals.py
---name: weekly-summarydescription: Use when writing the  weekly project summary.--- # Weekly summary 1. Read this week’s files in inputs/.2. Fill template.md, one part per stream.3. Match examples/good-summary.md.4. Run scripts/check_totals.py.5. Save to outputs/; list top risks.

Lines 2–4 are always visible. Keep SKILL.md under 500 lines; move detail into supporting files.

The Ascent · Learn · OrchestrateNext 06 / 13
The AscentL4 · Orchestrate · Build a skill

06 Skills

Build a skill for your team

Turn a task you already do well into a skill anyone on the team can run.

  1. Step 01

    Spot the repeat

    Work you do the same way every time

    • Weekly summary
    • Budget variance
    • Board pack
  2. Step 02

    Capture it

    Use the skill-creator skill if you have it (built in on claude.ai), or write SKILL.md by hand

    • give a real example
    • add your template
    • say what “good” is
  3. Step 03

    Test on real work

    Run it on 3 real jobs

    • fix its mistakes
    • tune the description
  4. Step 04

    Share it

    Put it where others can use it

    • the project’s .claude/skills/
    • your team’s shared library
Good skill checklist
  • One job
  • A clear “use when…” description
  • A real template and example
  • SKILL.md under 500 lines
Agents are built the same way: one Markdown file in .claude/agents/ — the specialist, its tools, what to report back.
The Ascent · Learn · OrchestrateNext 07 / 13
The AscentL4 · Orchestrate · Rules need checks

07 Instructions aren’t enforcement

A rule with no check is a wish

Toggle a gate off, then run the fleet. Watch which rules hold.

  • Instructions shape

    Rules files, room briefs and skills steer — agents can still drift.

  • Checks force

    Tests, scans and the human gate pass, fail and block.

  • Give a key rule both

    Write it down, and give it a check or a gate.

AGENTS.md

Charts read live data, never mock files

idle

No secrets committed

idle

Every figure traces to a source

idle

Toggle a gate off and run: the ungated rule slips straight through. Written policy shapes; only a check or the human gate forces.

The Ascent · Learn · OrchestrateNext 08 / 13
The AscentL4 · Orchestrate · Three ideas

08 What holds the level up

Three ideas carry the level

You govern the system; you don’t operate it keystroke by keystroke.

The rules are the work

A vague spec to four parallel agents gets four different systems, fast.

Write it before you dispatch.

Tempo is a governing move

Decide what runs at once and what holds: data first, builder waits, reviewer last.

Set the order of play.

Verify the action, not the report

A green build proves code ran. “Ready to merge” is a claim. Read the diff.

Trace the figure to its data.

The classic catch: an invented “+38% net sentiment” passes every check and ships to an exec dashboard. Only a human tracing it to the data stops it.
The Ascent · Learn · OrchestrateNext 09 / 13
The AscentL4 · Orchestrate · Do it now

09 Do it now

Run a small team, in real Claude

  1. Step 1 of 4

    Hire your GM

    You brief one lead; it runs the specialists.

    You should see: A task plan awaiting your go.

    paste into Claude
    You're my General Manager. I want a one-page plan for [a small project you care about]. Break it into 3 specialist tasks — researcher, writer, reviewer — with a one-line brief each. Don't start yet.
    Open Claude

  2. Step 2 of 4

    Dispatch a specialist

    Each specialist works its own brief.

    You should see: Focused output in that specialist’s voice.

    paste into Claude
    Run the researcher's task now, as that specialist. Return only their deliverable.
    Open Claude

  3. Step 3 of 4

    Assemble and check

    The GM integrates — and you watch for conflicts.

    You should see: One coherent deliverable, conflicts surfaced.

    paste into Claude
    Run the remaining specialists in order, then assemble everything into the finished piece. Flag anything where one specialist contradicts another.
    Open Claude

  4. Step 4 of 4

    Graduate

    On a computer, Claude Code runs these specialists in parallel for real — in rooms, guided by a CLAUDE.md that can import AGENTS.md.

    You should see: You, directing a department.

    When you’re at a desk: claude.com/claude-code — the same pattern with real folders, real parallel agents, and skills you invoke as /commands.

One step = one paste into Claude (the mobile app is perfect). Each step you mark done climbs a floor; progress saves on this device.

The Ascent · Learn · OrchestrateNext 10 / 13
The AscentL4 · Orchestrate · Right level?

10 Decide

Is this worth a team?

The Standing-Rules Test: is this a big, parallel job worth standing up rules and gates for?

Stay here — Level 4. Standing rules and gates pay for themselves.
Step down to Level 3 — Delegate. A fleet adds coordination cost with no parallel payoff.
Open Delegate
Do it yourself or Delegate — orchestration here is pure ceremony.
Open Delegate

Pick the closest match — you’ll get a verdict.

The Ascent · Learn · OrchestrateNext 11 / 13
The AscentL4 · Orchestrate · Capstone

11 Capstone

One full run, coached

  1. Pick a job that spans more than one area and that you’ll run again. Paste its data where marked.
  2. Write the spec and rules before the GM dispatches anything.
  3. Find the planted defect yourself, then fold 1–3 new rules back into the file.
No dev setup? Do every governing step on paper and let one chat role-play the fleet — the governing is the skill.
orchestration coach
You are my orchestration coach AND, separately, the lead agent (the GM) for a practice run. I want to learn to govern a fleet, not do the work myself.

The job: build a small dashboard/report from this input: [PASTE a CSV, dataset, or describe a real cross-area task you'll repeat].

Run it in this order and STOP at each numbered step for my input — don't race ahead:

1. Coach me to write a tight spec — the one outcome, the inputs, the concrete output, and "done means" — and push back if it's vague enough that four agents could drift.
2. Coach me to write a rules file (AGENTS.md/CLAUDE.md) with how-we-work, the commands, one HARD gate, and one machine-checkable acceptance criterion per room.
3. As the GM, decompose the job into specialist subagents and assign each a room. Tell me what runs in parallel vs. what HOLDS on a dependency. Pause so I can approve or redirect the decomposition and the order.
4. Role-play the fleet: report back as if each specialist worked in parallel and produced a branch. Deliberately include ONE plausible-but-invented or out-of-scope choice (a number with no source, a smoothing nobody asked for, a cross-room edit) so I can practice catching it. Do NOT tell me which.
5. Give me a check report mapped to my acceptance criteria, then wait while I review. I'll name the load-bearing claims, say how I'd verify each against the source myself, and identify the planted defect. After I answer, reveal it and tell me whether re-asking you would have caught it (it wouldn't).
6. Help me write the gate decision (merge / redirect / reject per branch) and the 1-3 new rules to fold back into the rules file so the next run starts smarter.

Throughout, hold me to governing before dispatching, setting policy and tempo not keystrokes, and verifying judgment rather than trusting green checks. Be blunt when I slip into the loop or rubber-stamp a green report.
Open Claude
The Ascent · Learn · OrchestrateNext 12 / 13
The Ascent · Learn

Wrap-up · check yourself

You can run a room

Self-check0 / 4

Remember

The rules are the work — every agent reads them first.
House rules always; a skill for one job; a subagent for a big one.
A rule with no check is a wish — give it a wall.
Verify the action, not the report — read the diff.

Top of the ladder. The test: hand your spec and rules to a stranger and leave the room. If the work comes back right and the bad branch is caught at the gate, your governance is doing the work. Go run a room.

Back to the overview Full reference