Skill: repo-autonomy

Purpose

Run end-to-end repo tasks with minimal user intervention while preserving safety, determinism, and reviewability.

Current Repo Mode

• •Model-first/offseason:
  • •Prioritize modeling improvements and offline validation loops.
  • •Defer non-blocking live reliability churn unless explicitly requested.

Use This Skill When

• •User asks for autonomous execution ("just do it", "handle everything", "run full flow").
• •User asks for fast merge readiness checks (lint/tests/e2e/PR updates).
• •User asks for real ingestion smoke tests.

Workflow

1. •Confirm active branch/worktree and use a dedicated worktree if substantial changes are needed.
2. •Gather only needed context (rg, targeted file reads).
3. •Implement with surgical edits.
4. •Validate:
   • •.venv/bin/ruff check .
   • •.venv/bin/pytest -q
   • •optional e2e run requested by user
5. •Summarize with exact outputs, unresolved risks, and next action.

Repo-Specific Safeguards

• •Prefer offline fixtures by default.
• •Real ingestion only on explicit request.
• •Generated data may be written under data/ only in active worktree and never auto-cleaned.
• •If dependency install changes versions (e.g., nfl_data_py), always rerun full tests.

Preferred Commands

• •Search/files: rg, rg --files
• •Lint: .venv/bin/ruff check .
• •Format: .venv/bin/black .
• •Tests: .venv/bin/pytest -q
• •Install deps: .venv/bin/pip install -r requirements.txt
• •Pipeline:
  • •.venv/bin/python -m pipeline.main --sport <sport> --stat <stat> --config <path> [--retrain]