---

name: repo-maintenance description: "Maintain a BayesianOptimization checkout with safe install, test, lint, docs, CI-matrix, and release-boundary guidance." disable-model-invocation: true metadata: disco-role: operating license: MIT

Repo Maintenance

Use this sub-skill only when the user is maintaining or editing a bayesian-optimization source checkout. It covers development installs, focused native tests, lint/format decisions, docs/notebook checks, CI matrix expectations, and safe debugging paths.

For package usage questions, route away from maintenance:

• Optimizer construction, registration/probing, fitting, save/load, or prediction workflows: ../optimizer-workflows/SKILL.md
• Acquisition-function selection or exploration/exploitation controls: ../acquisition-control/SKILL.md
• Constraints, typed parameters, categorical/integer domains, sequential domain reduction, or advanced examples: ../advanced-domain-features/SKILL.md

Maintainer workflow

1. Identify touched checkout files and choose the smallest native checks from references/development-workflows.md.
2. Prefer non-mutating verification first: focused pytest, Ruff format check, and Ruff lint check.
3. Use scripts/select_native_checks.py to print recommended command groups from touched paths or capability flags. The helper only prints commands; it does not execute them.
4. Escalate to full unit, notebook, docs, or CI-matrix coverage only when the touched files affect those surfaces or when preparing a broad change.
5. Use references/troubleshooting.md for Python/NumPy marker conflicts, pre-commit or Ruff failures, notebook timeouts, docs build failures, import confusion, and release/publish limits.

Safety rules

• Treat commands as checkout-maintenance commands, not runtime user guidance.
• Do not run scripts/format.sh unless the user explicitly wants formatting or automatic fixes; it mutates source files.
• Do not use scripts/check_precommit.sh as a routine verification shortcut; it installs hooks before running all files. Prefer direct non-mutating Ruff or pre-commit commands when verification is enough.
• Never provide credentialed publishing steps. PyPI publishing is release-only and guarded by repository secrets.