YOLO Mode
YOLO Mode
Section titled “YOLO Mode”YOLO mode controls how much autonomy agents have during workflow execution. When enabled, agents work without asking for validation at each step — like a senior team that you trust to get things done.
Three levels of autonomy
Section titled “Three levels of autonomy”| Level | Activation | Stops for | Best for |
|---|---|---|---|
| YOLO | yolo: true in .assemble.yaml | Production actions, missing info, external effects | Daily development |
| Hardcore | /yolo-hardcore command | Production only | Trusted environments, fast iteration |
| Full | /yolo-full command | Nothing | Maximum speed, accepts all risk |
Level 1 — YOLO (standard)
Section titled “Level 1 — YOLO (standard)”The default autonomous mode. Agents execute without intermediate validation but pause when:
- An action would affect production (deploy, database migration, etc.)
- Required information is missing and can’t be inferred
- An action has external side effects (sending emails, API calls, etc.)
Configuration
Section titled “Configuration”yolo: trueWhat changes
Section titled “What changes”| Without YOLO | With YOLO |
|---|---|
| Jarvis asks permission at each step | Jarvis executes the full workflow |
| Each deliverable requires validation | Deliverables are produced continuously |
| User controls the pace | Agents work at maximum speed |
| More oversight | More speed |
Level 2 — Hardcore
Section titled “Level 2 — Hardcore”Escalation beyond standard YOLO. Agents work with near-full autonomy:
- Stop only for production-impacting actions
- Infer information that can reasonably be deduced from context
- Skip intermediate validations entirely
Activation
Section titled “Activation”/yolo-hardcoreThis is a runtime command — not configurable in .assemble.yaml. Only a human can activate it, never an agent.
What changes
Section titled “What changes”- Agents interpret ambiguous requirements instead of asking
- Decisions are made autonomously based on best practices
- Only truly irreversible production actions trigger a pause
Level 3 — Full
Section titled “Level 3 — Full”Maximum autonomy. Agents stop for nothing. Full trust, full speed, maximum risk.
Activation
Section titled “Activation”/yolo-fullThis is also runtime-only and human-activated only.
When to use it
Section titled “When to use it”- Prototyping where speed matters more than perfection
- Sandbox environments with no production risk
- When you want to see what the team produces without any guardrails
Safety constraints
Section titled “Safety constraints”Regardless of YOLO level, these rules are non-negotiable:
- Hardcore and Full can only be activated by typing the command directly — no agent can activate them for you
- Agents must still follow their anti-patterns — YOLO doesn’t override quality rules
- The manifest is always updated — you can review what happened after the fact
- Deliverables are always produced in
assemble-output/— nothing is lost
Combining with governance
Section titled “Combining with governance”YOLO mode and governance are complementary:
| Governance | YOLO off | YOLO on |
|---|---|---|
| none | Manual validation at each step | Full autonomous execution |
| standard | Decision gates at phase transitions | Autonomous within phases, gates between |
| strict | Full audit trail + gates | Autonomous within phases, strict gates + audit |
With governance: strict + yolo: true, agents work autonomously within each phase but must still pass through formal decision gates between phases.
Recommended configurations
Section titled “Recommended configurations”Solo developer / prototyping
Section titled “Solo developer / prototyping”yolo: truegovernance: noneSmall team / startup
Section titled “Small team / startup”yolo: truegovernance: standardEnterprise / regulated
Section titled “Enterprise / regulated”yolo: falsegovernance: strict