Implements the autoresearch entry point requested in #907: - `SystemExperiment` class in autoresearch.py encapsulates the hypothesis → edit → tox → evaluate → commit/revert loop with configurable target, metric, and budget - `timmy learn` CLI command with --target, --metric, --budget, --max-experiments, --dry-run, --tox-env, and --model flags - Helper functions _extract_pass_rate() and _extract_coverage() for built-in metric extraction from tox/pytest output - program.md template in repo root for research direction - Unit tests for SystemExperiment and the learn CLI command Fixes #907 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
625 B
625 B
Research Direction
This file guides the timmy learn autoresearch loop. Edit it to focus
autonomous experiments on a specific goal.
Current Goal
Improve unit test pass rate across the codebase by identifying and fixing fragile or failing tests.
Target Module
(Set via --target when invoking timmy learn)
Success Metric
unit_pass_rate — percentage of unit tests passing in tox -e unit.
Notes
- Experiments run one at a time; each is time-boxed by
--budget. - Improvements are committed automatically; regressions are reverted.
- Use
--dry-runto preview hypotheses without making changes.