Project 02 · Minimal Harness
Objective
Build the same capability under controlled conditions and observe how harness design changes reliability.
Steps
- Define the task and acceptance criteria.
- Run the agent without a harness and record scope, tests, modified files and completion claims.
- Add AGENTS.md, feature_list.json and progress.md.
- Run the governed session with the same core prompt.
- Compare baseline vs harnessed execution.
Metrics
| Metric | Baseline | Harnessed |
|---|---|---|
| Files modified | ||
| Out-of-scope edits | ||
| Tests run automatically | ||
| Attempts to pass tests | ||
| Progress documented |
Next: Project 03