multi agent

Multi-Agent Collaboration

Andromity features a collaborative agent architecture where different specialist roles coordinate to design, write, test, and audit codebase improvements.

---

Agent Specialist Roles

Instead of a single "do-it-all" LLM prompt, Andromity uses targeted system prompt profiles that represent specialized agents:

1. Planner Agent: Performs deep research, examines file trees, runs search queries, and writes a detailed structural plan.
2. Coder Agent: Receives the approved plan, edits lines of code, and handles file creations.
3. Tester Agent: Executes build tasks and validation test suites, inspects console output, and provides feedback back to the Coder.
4. Reviewer Agent: Conducts a final audit, validating styling rules, checking for bugs, and presenting the final clean diff for Git.

---

Multi-Agent Workflows

In a typical developer session, these profiles operate sequentially to maximize accuracy:

1. User request: "Refactor user authentication schema."
2. Planner creates a step-by-step layout plan.
3. User reviews and approves the plan.
4. Coder edits the database schema files.
5. Tester executes pytest or vitest in the shell. If it fails, the Tester tells the Coder exactly why.
6. Reviewer audits the completed code and reports safety status.