Governor

An AI Development Control Layer — sits between the developer and the LLM, turning vague requests into controlled, risk-assessed engineering plans.

Status

v0.1-experiment: research seed. One command, fully tested offline pipeline.

Usage

py -m governor.cli plan <repo-path> "<change request>"

Runs four stages: ingest (repo scan) → analyze (risk) → decompose (subtasks) → promptgen (constrained prompts). Writes plan.json.

Use --kimi with KIMI_API_KEY set for real LLM analysis (default is an offline mock for development/testing).

Development

py -m pytest tests -v     # 11 tests, no API key needed

Research context

Implements the controlled workflow from: O. Hanan, "Controlled AI-Assisted Development in Student Software Projects" (AIFE 2026).