AEO is a Python-based engineering workflow layer for making AI-assisted software development measurable, reviewable, and progressively agentic.
V0.3 focuses on measurement integrity before autonomous agents are introduced.
- Crash-safe task finalization with a persisted finalization record
- Validation runs are linked to a task before subprocess execution
- A completed validation is reused after a partial failure instead of duplicated
- Failed validation keeps the task active so it can be fixed and retried
- Every validation attempt remains queryable for first-pass and retry analytics
- V0.2 task validation links are backfilled into the V0.3 additive schema
- Task history and task detail views
- First-pass validation success rate
- Validation retry rate
- Per-quality-gate pass/fail counts
- Per-quality-gate average duration
- Average run and task duration
Each new engineering run captures:
- AEO version
- Python version and implementation
- operating system and release
- machine architecture
- Git version
V0.3 does not modify existing V0.2 table columns. It adds:
execution_environmentstask_finalizationstask_validation_attempts
This allows an existing .aeo/aeo.db to be reused.
python -m uv sync --extra devKeep your existing .aeo/aeo.db if you are upgrading from V0.2.x.
python -m uv run aeo doctor
python -m uv run aeo task start "Implement feature X"
# work...
python -m uv run aeo task finish
python -m uv run aeo task history
python -m uv run aeo statsIf validation fails, the task remains active. Fix the reported issue and run task finish again; AEO records a new validation attempt.
python -m uv run aeo task history
python -m uv run aeo task show <TASK_ID>python -m uv run uvicorn aeo.api.main:app --reload --port 8009Endpoints:
GET /healthGET /runsGET /tasksGET /tasks/{task_id}GET /statsGET /docs
V0.4 is planned as the first Repo Guardian layer: deterministic repository intelligence and automated remediation around the quality system built in V0.1–V0.3.