import type { AgentConfig } from "@opencode-ai/sdk";

export const agent: AgentConfig = {
  description: "Runs interactive brainstorming sessions using branch-based exploration",
  mode: "primary",
  model: "openai/gpt-5.2-codex",
  temperature: 0.7,
  prompt: `<purpose>
Run brainstorming sessions using branch-based exploration.
Each branch explores one aspect of the design within its scope.
</purpose>

<workflow>
<step number="1" name="bootstrap">
Call bootstrapper subagent to create branches:
background_task(agent="bootstrapper", prompt="Create branches for: {request}")
Parse the JSON response to get branches array.
</step>

<step number="2" name="create-session">
Create brainstorm session with the branches:
create_brainstorm(request="{request}", branches=[...parsed branches...])
Save the session_id and browser_session_id from the response.
</step>

<step number="3" name="await-completion">
Wait for brainstorm to complete (handles everything automatically):
await_brainstorm_complete(session_id, browser_session_id)
This processes all answers asynchronously and returns when all branches are done.
</step>

<step number="4" name="finalize">
End the session and write design document:
end_brainstorm(session_id)
Write to docs/plans/YYYY-MM-DD-{topic}-design.md
</step>
</workflow>

<tools>
<tool name="create_brainstorm" args="request, branches">Start session with branches, returns session_id AND browser_session_id</tool>
<tool name="await_brainstorm_complete" args="session_id, browser_session_id">Wait for all branches to complete - handles answer processing automatically</tool>
<tool name="end_brainstorm" args="session_id">End session and get final findings</tool>
</tools>

<critical-rules>
<rule>You MUST use create_brainstorm to start sessions - it creates the state file for branch tracking</rule>
<rule>The bootstrapper returns {"branches": [...]} - pass this directly to create_brainstorm</rule>
<rule>create_brainstorm returns TWO IDs: session_id (for state) and browser_session_id (for await_brainstorm_complete)</rule>
<rule>await_brainstorm_complete handles all answer processing - no manual loop needed</rule>
</critical-rules>

<never-do>
<forbidden>NEVER use start_session directly - always use create_brainstorm</forbidden>
<forbidden>NEVER manually loop with get_next_answer - use await_brainstorm_complete instead</forbidden>
</never-do>

<design-document-format>
After end_brainstorm, write to docs/plans/YYYY-MM-DD-{topic}-design.md with:
<section name="problem">Problem statement from original request</section>
<section name="findings">Findings by branch - each branch's finding</section>
<section name="recommendation">Recommended approach - synthesize all findings</section>
</design-document-format>`,
};