syntaqxAn agent can turn on a project's g1t.page deployments, and have them from the moment it makes the repository: an agent (or Claude Code through the g1t MCP) could create a repository but not make it deploy, so flagon-io/session-lab needed a person to click Turn on deployments under Settings → Deployments, because only the site (that page, and Ask's new-repository card) called the deployments service's update_settings, and update_project's runs does not turn them on. Two operations now read and change a project's deployment settings as that page does, through the same service call: get_deployment_settings (GET /workspaces/{workspace}/projects/{project}/deployments/settings, deployments:read, the Read role as the service checks) and update_deployment_settings (PATCH there, repo:admin, the Admin role the page and the service require), in snake_case: enabled, production, previews, build_as, build_command, output_dir, dockerfile, instance_type, idle_days, via_workflow, with production_url, primary_domain, detected and framework read back. enabled true builds production at once; without the g1t plan it is refused with 402 payment_required saying where an owner starts it, without Admin with 403, and for a library, tool or other with 409, as the site says; bad values are refused with 422 before the service is asked. They are actions of the MCP workflow tool next to the deployment actions, and the CLI has g1t deploy settings (flags such as --previews off and --build-command change them), g1t deploy enable and g1t deploy disable, which find the current repository's project as g1t logs does (logs' resolve_project). create_repo takes deploy (REST, the repository tool's create, g1t repo create --deploy): once the repository is made its project's deployments are turned on as Ask's card does it, as its creator, and the response adds deployments (the settings) and deploy_error (why not); the repository is made either way, and deploy needs no scope beyond repo:create since the creator is its admin. Run credentials never change deployment settings (NEVER), and get_deployment_settings is a read for the audit log. The site keeps its own calls (ask.ts and the Settings page call the same service method the API now does). The deployments guide (Without the site, Settings), the deployments API guide, projects, the CLI guide, the MCP reference, reference.json and the regenerated openapi.json, CLI ops table and CLI reference say how. Deploy order: api, then the docs; the CLI commands ship with the next CLI release; no service or migration changes. Checked: cargo test -p g1t-api (211; openapi regenerated with G1T_WRITE_OPENAPI), -p g1t-contracts (316), -p g1t (110 + 51; ops_table and cli_reference regenerated with G1T_WRITE_OPS and G1T_WRITE_DOCS); cargo check -p g1t-api --target wasm32-unknown-unknown; node --test apps/web/app/lib/token-scopes.test.ts.56c1d3aHistory
