How enesmanan/gitarch works
What this project is (inferred from tree + minimal README)
GitArch appears to be a Python application that analyzes a Git repository and produces an “architecture/report” artifact, likely by:
• Cloning/loading a repo (via repo_manager.py)
• Building a structured context from the repo (context_builder.py, tools.py)
• Running a multi-step “agent” pipeline (agents/*, orchestrated by agents/orchestrator.py)
• Calling an external LLM provider (Gemini) (gemini_client.py)
• Presenting results via a web UI (presence of templates/index.html, templates/report.html, static/svg strongly suggests server-rendered HTML)
Request-driven UI flow (user triggers analysis from a web page)
2) Analysis pipeline (repo ingestion → context building → agent execution → report generation)
Orchestrator calls Repo Manager to fetch/open the repository.
5. Repo Manager reads/writes from Local Filesystem (and optionally hits Remote Git host if cloning).
Context Builder creates structured context from repo contents.
7. Orchestrator calls each Agent module (structure → overview → architecture → features → quality → summary).
Gemini Client calls Gemini API and returns responses.
10. Agents produce intermediate artifacts that feed into subsequent agents and ultimately into the final report.
Output → Templates → Report page
11. The final synthesized report is rendered via templates/report.html and served back to the browser.
Architectural patterns worth highlighting (specific to this structure)
• Pipeline / staged processing: multiple sequential analysis stages, each producing artifacts for the next.
• Strategy/Plugin-like “Agent” pattern: agents/base.py suggests a common interface; each agent is a replaceable module.
• Orchestrator pattern: central coordinator controlling flow and dependencies.
• Adapter pattern for external AI: gemini_client.py isolates API details from business logic.
Don’t overthink: capture “as many components as possible,” but keep truthfulness
Because the README is empty, avoid inventing specifics like “Redis cache” or “PostgreSQL” unless you confirm via requirements.txt or code.
But it *is* accurate and useful to:
• A generic “Web Framework runtime”
• “Local filesystem workspace”
• “External LLM API (Gemini)”
• Break agents into individual components (they exist in the tree)