Relevant source files

The following files were used as context for generating this wiki page: - [Traycer AI/plan_mode_tools.json](https://github.com/MKSK-Consulting/system-prompts-and-models-of-ai-tools/blob/main/Traycer%20AI/plan_mode_tools.json) - [Traycer AI/phase_mode_prompts.txt](https://github.com/MKSK-Consulting/system-prompts-and-models-of-ai-tools/blob/main/Traycer%20AI/phase_mode_prompts.txt) - [Google/Antigravity/planning-mode.txt](https://github.com/MKSK-Consulting/system-prompts-and-models-of-ai-tools/blob/main/Google/Antigravity/planning-mode.txt) - [Qoder/Quest Design.txt](https://github.com/MKSK-Consulting/system-prompts-and-models-of-ai-tools/blob/main/Qoder/Quest%20Design.txt) - [Amp/gpt-5.yaml](https://github.com/MKSK-Consulting/system-prompts-and-models-of-ai-tools/blob/main/Amp/gpt-5.yaml) - [Same.dev/Prompt.txt](https://github.com/MKSK-Consulting/system-prompts-and-models-of-ai-tools/blob/main/Same.dev/Prompt.txt)

# Traycer AI Traycer AI is an agentic AI system designed to function as a tech lead for an engineering team. Its primary role is to assist users by breaking down complex coding tasks and user queries into a series of high-level, structured phases. The system operates with read-only access to the codebase and focuses exclusively on coding-related aspects within the user's IDE, avoiding out-of-scope activities like deployment or account management. The core mechanic of Traycer AI involves analyzing a user's request, exploring the codebase using search tools, and then formulating a structured plan. This plan is communicated back to the user as a set of distinct phases. The AI is designed to seek clarification from the user when critical information is missing or pivotal design decisions are required, ensuring the final plan is well-aligned with the user's intent. *Sources: [Traycer AI/phase_mode_prompts.txt:3-8](), [Traycer AI/phase_mode_prompts.txt:18-21]()* ## Core Concepts Traycer AI's operation is built around the concepts of distinct operational modes or phases, structured task management, and a clear interaction flow for managing complex work. ### Operational Modes The agent operates in a cycle of distinct modes to manage a task from conception to completion. These modes ensure a structured approach to problem-solving, starting with research and planning, moving to implementation, and ending with verification. *Sources: [Google/Antigravity/planning-mode.txt:51-64]()* ```mermaid graph TD A[Start: User Query] --> B{PLANNING}; B -- Plan Approved --> C{EXECUTION}; C -- Implementation Done --> D{VERIFICATION}; D -- Verification Passed --> E[End: Task Complete]; C -- Unexpected Complexity --> B; D -- Minor Bugs Found --> C; D -- Major Flaws Found --> B; ``` This diagram illustrates the flow between Traycer AI's operational modes. * **PLANNING**: In this initial phase, the agent researches the codebase, understands the requirements of the user's query, and designs an approach. The primary output of this mode is a documented plan, which must be approved by the user before proceeding. If the user requests changes, the agent remains in PLANNING mode to update the plan. *Sources: [Google/Antigravity/planning-mode.txt:53-56]()* * **EXECUTION**: Once a plan is approved, the agent enters this mode to write code and implement the designed changes. If unexpected complexities arise that require a change in design, the agent can return to the PLANNING mode. *Sources: [Google/Antigravity/planning-mode.txt:58-59]()* * **VERIFICATION**: After implementation, the agent tests the changes to validate their correctness. Minor issues discovered during this phase can lead to a quick return to EXECUTION for fixes. However, if fundamental design flaws are found, the agent returns to the PLANNING mode to rethink the approach. *Sources: [Google/Antigravity/planning-mode.txt:61-64]()* ### Task and Plan Management Traycer AI utilizes a set of tools and artifacts to manage and track the progress of tasks. This includes breaking down work into checklists and formalizing the proposed changes for user review. A similar system, Qoder, also emphasizes creating a design document as a central artifact. *Sources: [Google/Antigravity/planning-mode.txt:69-95](), [Qoder/Quest Design.txt:71-77]()* | Artifact | Purpose | Source File | | ----------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------ | | `task.md` | A detailed checklist for organizing work. It tracks uncompleted, in-progress, and completed items. | `Google/Antigravity/planning-mode.txt` | | `implementation_plan.md`| A formal document created during PLANNING mode that outlines the goal, proposed changes, and verification plan for user approval. | `Google/Antigravity/planning-mode.txt` | | `walkthrough.md` | A summary document created after VERIFICATION, detailing the changes made, what was tested, and the validation results. | `Google/Antigravity/planning-mode.txt` | | `.same/todos.md` | A file used by the Same.dev agent to track progress with a list of todos that are updated as tasks are completed. | `Same.dev/Prompt.txt` | | `.qoder/quests/*.md` | A design document used by the Qoder agent to transform an idea into a high-level design. | `Qoder/Quest Design.txt` | ## Tooling Traycer AI is equipped with a specific set of tools to facilitate its workflow, primarily focused on planning and user interaction. Other agent systems provide context on similar tools for task management and user notification. *Sources: [Traycer AI/plan_mode_tools.json](), [Traycer AI/phase_mode_prompts.txt:13-14](), [Google/Antigravity/planning-mode.txt:26-49]()* ### Planning and Clarification Tools These tools are central to Traycer AI's primary function of breaking down tasks and interacting with the user. | Tool | Description | Parameters - | | `write_phases` | The primary tool for outputting the final plan. It breaks down the task into a list of phases, each with a title and a detailed description of the steps involved. This is used after the agent has achieved full clarity on the task. | **phases**: An array of phase objects, where each object contains:
- `title`: The title of the phase.
- `steps`: A list of strings describing the actions in that phase.
**summary**: A high-level summary of the entire plan.
**title**: The overall title for the plan. | | `ask_user_for_clarification` | Used to ask the user for more information when the query is ambiguous or a critical decision is needed. The agent can provide options to guide the user's response. | **question**: The question to ask the user.
**options**: An optional list of strings representing choices for the user. - | *Sources: [Traycer AI/plan_mode_tools.json](), [Traycer AI/phase_mode_prompts.txt:18-21]()* ### Task Boundary and Notification Tools These tools, described in a similar agent system, are used to manage the user's visibility into the agent's progress and to handle communication while a task is active. * **`task_boundary`**: This tool is used to enter a "task view" mode in the UI, giving the user a clear view of the agent's current task, summary, and status. It is called to signal the start of a major activity (like PLANNING or EXECUTION) and is updated as the agent progresses. *Sources: [Google/Antigravity/planning-mode.txt:26-42]()* * **`notify_user`**: This is the designated tool for communicating with the user while in "task view" mode. It is used to ask clarifying questions or to request a review of generated artifacts, such as an `implementation_plan.md`. Using this tool exits the task view mode. *Sources: [Google/Antigravity/planning-mode.txt:44-49]()* ## Interaction Flow The typical interaction with Traycer AI follows a structured sequence, ensuring that a task is well-understood and planned before any action is taken. This sequence diagram shows the high-level interaction flow from user query to the final plan. ```mermaid sequenceDiagram participant User participant Traycer_AI as Traycer AI participant Codebase User->>Traycer_AI: Submits activate Traycer_AI Note over Traycer_AI, Codebase: 1. Understand the query and codebase Traycer_AI->>Codebase: Use search tools extensively Note over Traycer_AI, User: 2. Seek clarification if needed alt Is clarification required? Traycer_AI->>User: ask_user_for_clarification(question, options) User-->>Traycer_AI: Provides clarification end Note over Traycer_AI: 3. Formulate and output the plan Traycer_AI->>Traycer_AI: Break down task into phases Traycer_AI-->>User: write_phases(title, summary, phases) deactivate Traycer_AI ``` *Sources: [Traycer AI/phase_mode_prompts.txt:16-21](), [Traycer AI/plan_mode_tools.json]()* The process is as follows: 1. The user initiates the process with a query. 2. Traycer AI begins by using its search tools to gain a comprehensive understanding of the user's request and the relevant parts of the codebase. 3. If there is any ambiguity or if a key decision needs to be made, the agent uses the `ask_user_for_clarification` tool. It will wait for the user's input before proceeding. 4. Once the agent has complete clarity, it structures the task into a series of high-level phases and uses the `write_phases` tool to present this plan to the user. *Sources: [Traycer AI/phase_mode_prompts.txt:16-21]()*