Agent mode gives p0 full execution capabilities, allowing it to make changes to your codebase, run commands, and implement solutions directly. This is the default mode for new sessions.
In Agent mode, p0 has access to all tools including:
- Bash - Execute shell commands
- Edit - Modify existing files with precise edits
- Write - Create new files or overwrite existing ones
- Read - View file contents
- Grep - Search for patterns in files
- Glob - Find files by name patterns
- WebSearch - Search the web for information
- WebFetch - Fetch content from URLs
- NotebookEdit - Modify Jupyter notebooks
- Task - Break down complex tasks
- Thinking - Internal reasoning
This full toolkit enables p0 to analyze your code and implement changes end-to-end.
How execution works
When you give p0 a task in Agent mode:
- The AI proposes tool calls (commands, file edits, etc.)
- Each tool call is executed in real-time
- Results and output are shown immediately
- The AI adapts based on the results and continues
You can see exactly what p0 is doing at each step, with full transparency into commands run and files modified.
Safety model
Agent mode can make real changes, but p0 keeps your main branch safe through worktree isolation:
- All changes happen in an isolated git worktree
- Your main working directory remains untouched
- You review diffs before merging changes back
- Worktrees can be discarded if you’re not satisfied
This isolation means you can experiment freely without risk to your current work.
Switching from Plan mode
If you start in Plan mode, you can switch to Agent mode after reviewing the plan:
- Approve the plan - p0 automatically switches to Agent mode and begins implementation
- Use /agent command - Manually switch to Agent mode at any time during the conversation
The conversation history is preserved when switching modes.
Agent mode can execute bash commands and modify files. Always review changes in the session diff before merging to ensure the implementation matches your expectations.
Next steps
