Agents, findings, and tools

Invoke an agent

Use the Agents panel to retain a specialist at Build home, or type a picker-authorized handle such as @ux-audit check this screen. Research, Strategy, Creative Direction, and authored agents answer in cited Chats using only their bounded context policy. Raw or pasted @text never gains execution authority.

Task Ask is informational. Task Run with a review agent creates a read-only review round against the exact worktree, diff, and Task context. Its structured findings support Discuss, Create task, Fix next round, Dismiss, and Accept risk. Review rounds never replace the implementation receipt or Changes.

Create a document

Research, Strategy, and Creative Direction can progressively generate a document inside their Chat. It begins as a draft: addressable from the message, but absent from Assets, search, MCP reads, project summaries, and downstream context. Save to Assets promotes the same document and makes it reusable project knowledge.

Create on Whiteboard

Supported actions can create Business Model Canvas, Product Bets, Personas, Positioning, Pitch Exploration, Journey Maps, Palettes, and Moodboards as Whiteboard page formats. The proposed page is appended without replacing collaborators' work.

Generate an image

Generate image uses specialized image models rather than the agent runtime. Each submission creates one file with prompt, model, ratio, references, and tool:image provenance, then saves it to Assets.

Availability

Agent availability is explicit. Missing, archived, revoked, disconnected, or fingerprint-changed agents fail without falling back to General. Connected-engine execution uses the user's configured engine; managed execution is an explicit credit-funded alternative. Unsupported review evidence is labelled Limited or Incomplete, never passed.