Traycer YOLO MODE: Plan, Sit & LET YOUR AGENT CODE! The REAL AGENTIC CODING!

Key Concepts

• Tracer: An AI tool for planning and verifying AI code.
• YOLO Mode: A new feature in Tracer that enables fully autonomous AI coding loops.
• You Only Look Once (YOLO): The principle behind YOLO mode, meaning the AI operates autonomously without constant user intervention.
• Custom CLI Agents: A feature allowing users to create personalized agents with specific permissions and flags for command-line tools.
• User Scope vs. Workspace Scope: Options for saving custom CLI agent configurations, either for individual use or for team collaboration.
• claude dangerously skip permissions flag: A specific flag used with CLI tools (like Claude) to bypass permission prompts for automation.
• $tracer_prompt variable: A required variable for Tracer to pass instructions to the CLI agent.
• Phase Breakdown: The process of breaking down a large task into smaller, manageable phases.
• Plan Handoff vs. User Query Handoff: Methods for passing requirements to the AI agent; plan handoff is recommended for better accuracy.
• Verification Handoff: Settings that define the exit criteria for the autonomous loop, specifying the severity of issues that will trigger a fix cycle.
• Critical, Major, Minor Issues: Categories of bugs used in verification handoff to determine if code needs to be re-processed.
• Agentic Workflow: A workflow where an AI acts as an autonomous agent, rather than requiring constant user direction.

YOLO Mode: Revolutionizing AI Code Automation in Tracer

This video introduces a significant update to the AI coding tool Tracer, named "YOLO mode," which transforms the user experience from manual oversight to full automation. Previously, Tracer excelled at planning and verifying AI code, but still required manual clicks for execution, verification, and bug fixing, making the user the bottleneck. YOLO mode addresses this by enabling a fully autonomous loop where the AI handles planning, coding, verification, and self-correction without user intervention.

Enabling True Automation: Custom CLI Agents

A crucial prerequisite for YOLO mode's seamless operation is the ability to bypass permission prompts from command-line interface (CLI) tools. These prompts, while good for safety, halt automation when the user is not present. Tracer's new "Custom CLI Agents" feature allows users to create personalized versions of their favorite tools with predefined permissions.

Setting up Custom CLI Agents:

1. Navigate to Settings or the Integrations menu in Tracer.
2. Locate and select the Custom CLI Agents section.
3. Click to Create a new one.
4. Choose the Scope:
   • User Scope: The agent configuration is saved for the individual user and travels across all projects.
   • Workspace Scope: The configuration is saved in a .tracer folder within the project root, allowing it to be committed to a repository for team-wide use, thus preventing "works on my machine" issues.
5. Name the Agent: For example, "Claude Code YOLO" to distinguish it from standard agents.
6. Define the Command: This is where automation is unlocked. For tools like Claude, instead of the standard command, use a special flag provided by the AI provider. For Claude, this is claude dangerously skip permissions $tracer_prompt.
   • The $tracer_prompt variable is essential as it's how Tracer passes its instructions to the CLI.
7. Save the configuration.

This setup creates an agent that can execute commands without stopping to ask for permission.

The YOLO Mode Workflow: From Planning to Autonomous Execution

The YOLO mode is activated from the Tracer task interface, specifically on the Kanban board view where phases are displayed.

Steps to Activate YOLO Mode:

1. Phase Breakdown: As in previous versions, users interact with Tracer to break down a task into phases. For instance, adding a watch list feature to a movie app, involving phases like setting up a local database schema, creating UI elements, and building a "My Watch List" screen.
2. Initiate YOLO Mode: On the Kanban board, click the new YOLO Mode button.
3. Configure YOLO Settings: A detailed configuration panel appears, offering granular control:
   • Phase Slider: Users can specify which phases to automate, either a subset or the entire project.
   • Agent Selection: Choose the custom CLI agent created earlier (e.g., "Claude Code YOLO").
   • Handoff Settings:
     • Plan Handoff (Recommended): Tracer generates a detailed, step-by-step architectural plan before handing it to the AI agent. This significantly reduces the chance of the AI deviating from the task.
     • User Query Handoff: The raw requirement is directly passed to the agent, which is faster but potentially less accurate.
   • Verification Handoff: This is the most critical setting, defining the exit criteria for the autonomous loop. Users can select checkboxes for Critical, Major, and Minor issues.
     • If Critical and Major issues are checked, Tracer will verify the code. If it finds syntax errors (critical) or logic bugs (major), it will automatically reject the code, create a fix plan, and send it back to the AI for correction.
     • Minor issues (e.g., suboptimal variable names, missing comments) will be ignored to prevent the loop from getting stuck on trivial matters.
4. Start Automation: Once configured, click Start Automation.

The Autonomous Loop in Action

With YOLO mode activated, Tracer takes over:

1. Planning: Tracer creates a detailed plan for the current phase.
2. Handoff to Agent: The plan is passed to the selected custom CLI agent (e.g., Claude).
3. Execution: The agent, using the dangerously skip permissions flag, executes the plan without user prompts. This includes installing libraries, writing code, and potentially running tests.
4. Verification: Once the agent finishes, Tracer automatically runs a verification scan based on the defined criteria (critical, major, minor issues).
5. Fixing (if necessary): If critical or major issues are found, Tracer automatically generates a fix plan and sends it back to the agent to iterate.
6. Loop Continuation: This loop of plan, handoff, verify, and fix continues until no critical issues remain.

Practical Considerations and Pro Tips

• Computer Sleep: To prevent the script from pausing, ensure your computer does not go to sleep during long YOLO sessions. Consider setting it to "never sleep" or using utility apps.
• Artifact Slots: YOLO mode consumes usage slots. Ensure sufficient capacity is available for the task to avoid pauses.
• Quality of Life Upgrade: The reduction in manual clicks is a significant improvement, shifting the user's role from operator to architect. Users define the phases, quality standards, and verification strictness, then let the AI agents work autonomously.

Conclusion

Tracer's YOLO mode, coupled with custom CLI agents, represents a substantial leap towards truly agentic workflows. It empowers users to delegate complex coding tasks to AI with unprecedented autonomy, allowing them to focus on higher-level architectural decisions. The ability to chain phases together and define strict verification criteria makes the process efficient and reliable, marking a significant advancement in AI-assisted software development.