Code Quality: Refactor agent_loop.py to a strict State Machine

[pradeep0153]

Closes #240.

Description

This PR executes a foundational architectural refactoring of the core logic within Repopilot. The central AutonomousAgent.run() method in agent_loop.py had grown into a massive procedural while-loop with deep nesting, complex continue/break conditions, and tangled error handling.

This procedural architecture made it incredibly brittle to add new features (like the recent --lint gate). The logic for repository ingestion, LLM interaction, sandbox execution, git committing, and rollback were all mashed together in a single monolithic function, violently violating the Single Responsibility Principle.

Changes Made

• Codebase Audit: Systematically profiled the monolithic loop structure in agent_loop.py.
• State Machine Pipeline: Refactored the core execution loop to utilize a strict State Machine pattern.
• Decoupling & Encapsulation: The architecture now mathematically defines explicit states: INGEST, PLAN, MODIFY, LINT, TEST, COMMIT, and ROLLBACK. Each state is encapsulated into its own distinct, isolated function/class.
• Execution: The main agent loop is now mathematically reduced to a simple state transition manager that passes context payloads between these distinct, pure functions based on explicit success/failure signals.

Impact

• Maintainability & Developer Velocity: Cramming complex logic into a single 300-line while-loop is a fatal anti-pattern that guarantees regressions and developer paralysis. By engineering this strict State Machine architecture, we mathematically decouple the agent lifecycle. This drastically accelerates maintainability, makes the core loop highly testable via unit tests, and allows us to seamlessly inject new middleware (like a security-scanning state) without touching existing logic.

[vercel]

@pradeep0153 is attempting to deploy a commit to the sreerevanth's projects Team on Vercel.

A member of the Team first needs to authorize it.