How a session works

Coframe Agent is an autonomous agent — you describe what you want, and it handles the rest. This page breaks down who does what at each stage of a session.

The overall flow

You describe intent → Agent plans → Agent builds → Agent tests → You review

That’s it. You set the direction; the agent does the work. You can watch, interact, and give feedback at any point, but you never have to manage tasks, write code, or run tests yourself.

Responsibilities breakdown

What you do

Stage	Your role
Start	Open a session and describe what you want in the chat.
Watch	Observe the agent’s progress — tasks, file changes, preview updates.
Preview	Interact with the live preview to see the change as a real user would.
Feedback	Send follow-up messages or add annotations to guide the agent.
Review	Check the final result and approve, request changes, or iterate.

What the agent does

Stage	Agent’s role
Requirements	Extracts concrete, testable requirements from your description.
Planning	Creates a task plan — ordered steps that map to requirements.
Implementation	Writes code, edits files, and applies changes to the workspace.
Preview	Updates the live preview so you can see results immediately.
Testing	Runs the test suite and maps results back to requirements.
Auto-fix	Detects failed tests and iterates to fix them automatically.
Delegation	Spawns sub-agents for independent tasks that can run in parallel.
Report	Produces a summary of what was done, what passed, and what needs attention.

You can always step in

The agent works autonomously, but you’re never locked out:

• Send a chat message at any point to redirect, clarify, or add context.
• Annotate the preview to point at specific elements and leave visual feedback.
• Watch sub-agents in the sub-agents panel to see how parallel work is progressing.

The agent treats your feedback as the highest-priority input. When you send a message or annotation, the agent will pause its current work, incorporate your feedback, and adjust its plan accordingly.

Example session

Here’s what a typical session looks like — you describe a change, the agent extracts requirements, plans tasks, implements, tests, and reports back. When you give feedback, the cycle repeats:

The cycle of describe → agent works → you review → give feedback repeats until you’re satisfied with the result.