/initDoc

Actions

0. STEP 0:

   • Obtain the current project structure.
   • Read key files, such as various README.md / package.json / go.mod / pyproject.toml ...

1. Step 1: Global Investigation (using scout)

   • Launch concurrent scout agents to explore the codebase and produce reports.

2. Step 2: Propose Core Concepts & Get User Selection

   • After scouting is complete, perform a synthesis step: Read all scout reports and generate a list of candidate core concepts (e.g., "Authentication", "Billing Engine", "API Gateway").
   • Use the AskUserQuestion tool to present this list to the user as a multiple-choice question: "I've analyzed the project and found these potential core concepts. Please select the ones you want to document now:".

3. Step 3: Generate Concise Foundational Documents

   • In parallel, launch dedicated recorder agents to create essential, project-wide documents.
   • Task for Recorder A (Project Overview): "Create overview/project-overview.md. Analyze all scout reports to define the project's purpose, primary function, and tech stack."
   • Task for Recorder B (Coding Conventions): "Create a concise reference/coding-conventions.md. Analyze project config files (.eslintrc, .prettierrc) and extract only the most important, high-level rules."
   • Mode: These recorders MUST operate in content-only mode.

4. Step 4: Document User-Selected Concepts

   • Based on the user's selection from Step 2, for each selected concept, concurrently invoke a recorder agent.
   • The prompt for this recorder will be highly specific to control scope and detail: "Task: Holistically document the <selected_concept_name>. 1. Read all relevant scout reports and source code... 2. Generate a small, hierarchical set of documents:
     • Optionally, create ONE overview document if the concept is large enough to require its own high-level summary (e.g., overview/authentication-overview.md).
     • Create 1-2 primary architecture documents. This is mandatory and should be the core 'LLM Retrieval Map'.
     • Create 1-2 primary guide documents that explain the most common workflow for this concept (e.g., how-to-authenticate-a-user.md).
     • Optionally, create 1-2 concise reference documents ONLY if there are critical, well-defined data structures or API specs. Do not create reference docs for minor details. 3. Operate in content-only mode."

5. Step 5: Final Indexing

   • After all recorder agents from both Step 3 and Step 4 have completed, invoke a single recorder in full mode to build the final index.md from scratch.

6. Step 6: Cleanup

   • Delete the temporary scout reports in /llmdoc/agent/.