Set up guardrails

1. Context: list_workspaces(), list_projects(workspace_id); list_actors(workspace_id), else create_actor(...).
2. One checklist per flow: create_checklist(...) with focused instructions and 3 to 6 crisp pass/fail checks. Small checklists give clearer failure signals.
3. Standards: list_project_guidelines(project_id), then add_project_guideline(...) for anything missing. Guidelines reach every run, so checks can assert them without restating them.
4. Green baseline first: run_checklist(checklist_id) for each and poll get_checklist_run(run_id). Fix flaky checks now; a noisy guardrail gets ignored.
5. Automate: create_workflow(...) with a trigger: schedule with { cron, timezone } in trigger_config (timezone from get_me; daily is a good default), or vercel_event / github_event for deploys and PRs when those connections exist (preferred; a schedule also catches content-only regressions).
6. Steps: add_workflow_step(workflow_id, step_type, config) appends and connects, so add in run order: one run_checklist step per checklist (config={ checklist_id }), then a notification step (config={ channel: "email", recipients: { type: "all_members" }, template: "failed" }).
7. Check and activate: get_workflow(workflow_id): execution_order must cover every step and unreachable_step_ids be empty. Then toggle_workflow(workflow_id).
8. Tell the user what is covered, when it runs, where failures go, and how to extend it. Everything is manageable at app.stuntdouble.io.