Repository navigation
Autonomous Execution
github-actions[bot] edited this page Feb 28, 2026
·
1 revision
Autonomous Execution lets ClawDE complete multi-step tasks without step-by-step approval — you set a goal, review the plan, and the AI executes until done or blocked.
- Set a session to Autonomous mode (toggle in session header or via
clawd session --mode autonomous) - Describe the goal: "Migrate all database queries from raw SQL to the ORM"
- The AI creates a plan (visible in the plan panel) and begins executing tasks
- Execution continues until the goal is complete, a blocking error occurs, or context runs out
- You review a single consolidated diff at the end
In Autonomous mode, the daemon approves:
- File reads
- File writes that match the stated goal
- Shell commands that are read-only (
ls,cat,cargo test, etc.)
Even in Autonomous mode, these always pause for approval:
- Destructive shell commands (
rm,git reset,DROP TABLE) - Network requests outside localhost
- Commits and pushes
- Changes outside the stated repo scope
You can pause Autonomous execution at any time from the session header. The AI saves its state (current subtask, completed steps, pending diffs) and waits. Resume picks up from exactly where it left off.
| Autonomous Execution | Builder Mode |
|---|---|
| Single session | Multi-session graph |
| Sequential by default | Parallel subtasks |
| Goal → execute | Goal → plan review → execute |
| Best for focused tasks | Best for large feature work |
Planned for Sprint JJ alongside Builder Mode. Requires the task-graph engine and a session-level approval policy API.
ClawDE · GitHub · MIT License
ClawDE
Getting started
Reference
- Architecture
- Daemon-Reference
- Folder-Structure
- Providers
- Multi-Account
- Security
- Features
- Features/Daemon
- Features/Session-Manager
- Features/Mode-System
- Features/Repo-Intelligence
- Features/Projects
- Features/Remote-Access
- Features/Desktop-App
- Features/Provider-Knowledge
- Features/Coding-Standards
- Roadmap
Branding
Contributing