stuntdouble-basics

stuntdouble-basics:

• Stunt Double runs AI actors (user personas) that drive real browsers through products and return evidence. Auth is OAuth 2.1 with PKCE; no API keys.
• Start with list_workspaces, then list_projects; always pass a workspace. Anything with a reachable URL can be tested (production, staging, previews, published prototypes).
• search finds anything in a workspace by name or content (use types and project_id to narrow). Search before you create, and prefer it over listing and filtering. It only returns workspaces you belong to.

Actors:

• Name like "FirstName - Role", give a description and system prompt, and cover varied segments (new, power, enterprise, accessibility, mobile). add_actor_knowledge is for what one actor needs to remember.
• Conversations are read-only over MCP (list_conversations, get_conversation). To probe actors for research or design review, run an interview.

Workflows and checklists:

• Workflows validate multi-step journeys; checklists are point-in-time quality checks. Runs are async: poll get_workflow_run or get_checklist_run.
• create_workflow (trigger_type "schedule" or "webhook"), then add_workflow_step once per step in run order; a workflow with no steps does nothing. Steps run along their connections: check execution_order and unreachable_step_ids from get_workflow before activating.
• Schedules need a timezone: pass { cron, timezone } in trigger_config using the timezone from get_me, or the run lands at that hour in UTC.

Feedback:

• Lifecycle: new, reviewed, resolved (or dismissed for false positives). Cross-reference with actors and workflows for patterns and coverage gaps.

Interviews:

• Build the guide (add_interview_section, add_interview_item with question or task) before launch_interview. Add participants with add_interview_participant (actor_id, or persona_spec with name, bio, traits).
• Launch is async; follow with get_interview and get_interview_participant, read get_interview_report, and use regenerate_interview_report after transcripts change.

Guidelines (workspace standards):

• A guideline is a standing rule (design system, tone of voice, brand, accessibility, compliance and so on) owned by the workspace and attached to projects. Enabled rules are rendered into every checklist run, design review, interview and triage for the project, so read list_project_guidelines first and assert the rules rather than restating them.
• Check list_workspace_guidelines, then write once with add_project_guideline (library plus this project) or add_workspace_guideline (library only), and reuse elsewhere with set_project_guideline. update_workspace_guideline with apply_to_design_reviews also covers Slack and Linear reviews, which have no project.
• To enforce a standard: codify it as a guideline, assert it with checklist checks a user could verify on the rendered page, then re-run via create_workflow on a schedule or deploy and design events.

Stunt Double Index (how AI agents experience websites):

• For agent readiness or AI visibility questions use get_index_report (a domain, or a project_id), list_index_sessions to explain a score, and search_index_domains to compare. Index data is public.
• For a site the Index does not track yet, request_index_score adds it with a first score (light probes; include_agent_sessions for live sessions). It never rescores a tracked site.
• request_index_rerun spends about 24 agent sessions, is owner-only and allowed once every 10 minutes: use it after shipping changes, not to refresh a current report.

Scope:

• get_workspace settings (public sharing, widget, workers, network policy) are admin ceilings a project cannot override.
• Projects are archived, never deleted; archived projects and their contents read as missing.
• list_project_mcp_servers shows which MCP servers a project's runs can reach.

Prompts: the server ships recipes, usually surfaced as slash commands: validate_design, verify_change, run_user_research, triage_feedback, setup_guardrails, check_brand, check_design_system, check_compliance, check_continuity, stuntdouble_guide.