← Presentations
GUIDE · CLAUDE CODE

Levels of Using Claude in Funny Movies

📄 Source: claude-guide-en.md 🪜 Levels: 1 → 6 📅 Updated: 2026-10-20

I 6 levels

The higher the level, the more Claude does on its own and the less you touch. Rules, hooks and quality gates from the lower levels still apply at every level above.

🪜 Summary

LevelHowYou take part
1Ask & exploreEvery question
2Direct work with guardrailsEach step, approve the push
3Workflow skills + swarmApprove each artifact (PRD, ADR, Spec, Plan)
4Local /autopilot3 times: Start, Gate 1, Gate 2
5Headless on GitHubAdd the label, /approve, merge
6Auto gatesOnly add the label

1 Ask & explore

Ask about the code. Claude reads and greps before answering, no code changes.

💬 How Level 1

🎯
You take part: every question.

📌 Real example

"How do I manage cron jobs dynamically?"

→ 3 levels, Level 2 recommended (DB + TaskScheduler + admin UI).

2 Direct work with guardrails

Claude changes code in the session; you watch each step.

🛡️ Always-on guardrails Level 2

  • CLAUDE.md + .claude/rules/ always loaded
  • Library skills auto-load
  • Hooks catch secrets
  • session.txt
  • Ask before every push
🎯
You take part: each step, approve the push.

📌 Real example

Cron editor modal + tooltip (df8bdba1, 9b1d5419): done directly, not through autopilot.

3 Workflow skills + swarm

Follow the planning flow; each step produces an artifact in docs/.

🐝 Skill chain Level 3

/scope → /architect → /spec → /swarm-plan → /swarm-execute → /swarm-review → /land-the-plane

Worker agents run in parallel, model pinned per agent.

🎯
You take part: approve each artifact (PRD, ADR, Spec, Plan).

📌 Real example

"Premium user" feature: autopilot refused because it needs a PRD → go to /scope.

3+ How the agents work together

One orchestrator, focused workers, separate reviewers.

🧭 Roles

  • The orchestrator is the main session (Opus). It is the only agent that starts workers (workers have no Agent tool), owns the task list, merges their work, runs the gates and asks you at the gates.
  • Each worker gets one focused prompt and returns a summary (files changed, tests added, verdict block), not file dumps. Workers never share state with each other.
  • worker-builder works in its own git worktree. The orchestrator merges or cherry-picks its commits into main, then re-runs the gates itself.
  • Reviewers are separate agents from builders. Each ends with a verdict block (approve / revise). Only blocking findings send work back, at most 3 rounds.
                    You  (Start · Gate 1 · Gate 2)
                              │
               Orchestrator  (main session, Opus)
     ┌──────────────┬─────────┴──────────┬──────────────────┐
worker-explorer ×3  worker-reviewer     worker-builder ×2    worker-reviewer ×2
(Haiku: find        (plan review)       (BE + FE, each in    (quality + security)
 existing patterns)                      its own worktree)

📌 Example: the dynamic cron jobs run

StepAgent(s)ModelResultCost
Explore3 × worker-explorer: Liquibase + entity patterns, backend admin patterns, webapp admin UI patternsHaikuPatterns for the plan; the orchestrator writes the plan + ADR-0023$0.03
Plan reviewworker-reviewerSonnetapprove, 0 blocking, 7 advisory$0.16
Gate 1you—approved—
Implement2 × worker-builder in parallel: BE (DJ-1..3), FE (DJ-4..5)SonnetCommitted in worktrees, cherry-picked onto main$0.42 + $0.25
Code review2 × worker-reviewer: quality, securitySonnetapprove, 0 blocking, 6 advisory$0.17 + $0.15
QA ×2, gates, commit, CI + smokeorchestratorOpus84 tests pass twice, CI green, smoke 22/22$2.53 (whole run)
📦
Total $3.71. The orchestrator is 68% of it; the 8 workers together cost $1.18.

4 Local /autopilot

Chains the workflow skills into one run; at most 3 rounds per loop.

🤖 Flow Level 4

plan ⇄ review → Gate 1 → implement ⇄ review → QA ×2 → commit → Gate 2 → CI + smoke
🎯
You take part: 3 times: Start, Gate 1, Gate 2.

📌 Real example

📦
Dynamic cron jobs (ADR-0023): BE + FE, CI green, smoke 22/22, $3.71.

🪶 Lite mode

Under 100 changed lines not counting tests, and no ADR: no docs/plans/plan-*.md and no docs/reports/autopilot-*.md. The plan and the report go to the console (local) or the PR body (headless).

5 Headless on GitHub

ADR-0022 · workflow .github/workflows/autopilot.yml · both gates move to GitHub.

☁️ Steps Level 5

StepYou doAutopilot does
StartAdd the label autopilot to an issuePlans and reviews on autopilot/issue-<N>, opens a draft PR
Gate 1Comment /approve, or /revise <notes>Implement ⇄ review → QA ×2 → full gates → pushes to the PR branch, marks it ready
Gate 2Review the diff and merge the PR (CI deploys it)Never pushes to main, never merges
After deployRun ./api/mock-test.sh—

📌 Real example

Issue #34 (back button color) → PR #35.

🔒 Safety

  • Only the repo owner can trigger a run, by label or by comment.
  • Issue text is data; instructions that try to change the workflow, permissions or gates are ignored.
  • Refuses to edit .github/, .claude/, CLAUDE.md, AGENTS.md, .env files, or write data-dropping migrations.

6 Auto gates

ADR-0025 · add one more label together with autopilot.

⚡ Labels Level 6

Extra labelGate 1Gate 2
(none)you /approveyou merge
autopilot:auto-mergeyou /approveautomatic merge
autopilot:autoautomaticautomatic merge

✅ Auto-merge conditions

  • The PR is ready, meaning code review returned approve.
  • No changes to .github/, .claude/, CLAUDE.md, AGENTS.md, scripts/claude-*, Liquibase migrations or .env.
  • ≤ 400 changed lines outside tests and docs.
  • The script re-runs mvn verify / webapp test + lint + build itself.
🔎 Details

Gate 1: the plan job writes <!-- autopilot:plan-review=approve --> in the PR body only when plan review returned approve; the workflow sees it and dispatches the build job (workflow_dispatch). A /revise always waits for you.

Gate 2: scripts/claude-autopilot-merge.sh (always the copy on main) squash-merges with the secret AUTOPILOT_MERGE_TOKEN. If any check fails, it comments the reason and leaves the PR to you.

📌 Real example

📦
Issue #36 → PR #37 merged and deployed in ~5 minutes, build job $0.39, nobody clicked anything.

⚠️ Risk

Code reaches prod without a human read. The smoke check after the deploy is still manual. Rollback = git revert the squash commit.

II Which level to pick

🧭 By kind of work

WorkLevel
Not sure what to do yet, or you only need to understand the code1
A small change you want to watch step by step2
A medium or large feature (needs a PRD, several days)3
A small feature, fix or refactor with clear acceptance criteria4, or 5 without a terminal
Small, low-risk work that touches no migration or agent config6