Apply

Recommendations are read-only.
Apply is explicit.

xops can see useful missing capabilities, but seeing is not mutating. When xops can add something, it first builds a setup plan with packages, files, commands, risks, approvals, and rollback hints. Nothing changes until the user approves.

Example — applying Newman recommendation

Show the plan.
Then ask.

xops recommendations apply rec_postman_newman_runner

setup plan
This will: - install dev dependency: newman - modify package.json - update the lockfile - add script: test:api - optionally add a CI job if selected Risks: - adds a dependency - changes package metadata - may affect CI runtime Approval required: yes

Example — applying Renovate recommendation

Local config only.
No external account.

xops recommendations apply rec_dependency_monitoring_renovate

setup plan
This will: - create renovate.json - use the detected package manager/workspace layout - group updates based on repo structure - not install a GitHub app - not create external credentials Approval required: yes

Example — applying MaGit recommendation

Init, never deploy.
Nothing touches production.

xops recommendations apply rec_magit_metadata_repo

setup plan
This will: - install or add @x12i/magit if required - run MaGit init only after approval - create .magit local state/config - not deploy anything - not apply anything to production Approval required: yes

Agent-safe setup

Agents can propose setup.
Humans approve mutation.

A code agent should be able to discover that a useful supported capability is missing. It should also be able to ask xops for the setup plan. But mutation still needs control — in agent mode, xops returns the plan and exits with approval required unless --yes is explicitly provided.

$ xops recommendations apply rec_postman_newman_runner --dry-run --json --agent → install newman, modify package.json, update lockfile, add test:api, approval required $ xops recommendations apply rec_postman_newman_runner --yes ✓ executes — approval given explicitly