External Lib Flow

Onboarding Routine - Execute sequentially - Use Todo Tasks

Context

• Purpose: Onboard AI to external codebase for usage understanding
• Tool: Repomix (MCP or CLI) to package codebase (compressed XML)
• Target: Two documents in refsrc folder and in Rosetta
  1. File: {project-name}.xml (compressed codebase, unmodified Repomix output)
  2. File: {project-name}-onboarding.md (brief Learning Flow with reference)
• Files involved: Project path, README, package files for auto-detection
• MUST generate brief Learning Flow (3-5 words per step, max 20 lines)
• MUST use compressed XML (Tree-sitter) for small output
• Onboarding document MUST specify KB title and search instructions
• Update ARCHITECTURE.md based on template MUST use refsrc/{project-name}.xml and refsrc/{project-name}-onboarding.md. MUST use grep or search with those, because those are big files.. Combine this rule for multiple external dependencies.
• Workflow state MUST be saved to agents/TEMP/<FEATURE>/external-lib-flow-state.md file.

Onboarding Flow

Phase 0: Prerequsites

1. All Rosetta prep steps MUST be FULLY completed
2. USE SKILL load-project-context, orchestration, hitl
3. MUST ALWAYS use todo tasks ledger, ASAP. Phases are sequential. Independent tasks can run in parallel.

Phase 1: Discovery

1. Ask project path
2. Auto-detect project name
3. Auto-detect version number
4. Auto-detect tech stack

Phase 2: Analysis

1. Package with compression enabled
2. Analyze README for usage
3. Extract main entry points
4. Generate Learning Flow summary

Phase 3: Publishing

1. Upload compressed XML as-is
2. Create brief onboarding document
3. Confirm both document IDs
4. Cleanup temporary files

Phase 4: Verification

1. Search by project name
2. Verify tags present
3. Display Learning Flow
4. Confirm AI onboarded

Make sure to have todo tasks for each step! Do not skip steps!

Phase 1: Discovery

Idea

Ask or confirm user for project path with helpful suggestions. Auto-detect all metadata from project files to minimize user questions.

Key Points

• Only ONE question: project path
• If user forgets, suggest current directory, parent, common paths
• Auto-detect project name from directory or package files
• Auto-detect version from package.json, pyproject.toml, pom.xml, Cargo.toml, etc.
• Auto-detect tech stack from file extensions and package files
• Use directory name as fallback for project name

Steps

1. Ask user: "What project path to onboard?"
   • Suggest existing project names with relative paths, which could be potentially shared
2. Validate path exists and is accessible
3. Auto-detect project name:
   • Check package.json (name field)
   • Check pyproject.toml (name field)
   • Check pom.xml (artifactId)
   • Check Cargo.toml (name field)
   • Any other project files (*.csproj, etc.)
   • Fallback: Use directory name (last path segment)
4. Auto-detect version:
   • Check package.json (version)
   • Check pyproject.toml (version)
   • Check pom.xml (version)
   • Check Cargo.toml (version)
   • Any other project files (*.csproj, etc.)
   • Fallback: skip version tag
5. Auto-detect tech stack:
   • package.json → ["nodejs", "javascript"] + scan for "typescript", "react", etc.
   • pyproject.toml or requirements.txt → ["python"] + scan for frameworks
   • pom.xml → ["java", "maven"]
   • Cargo.toml → ["rust", "cargo"]
   • Any other project files (*.csproj, etc.)
   • Multiple detected → include all
6. Inform user of detected metadata: "{project-name} v{version}, tech: {tags}"