Design Conversations

A design conversation is Trurlic’s interactive mode. There is no trurlic command that runs it — the conversation happens through your coding agent, over MCP. The agent asks probing questions, you reason through the tradeoffs together, and each decision you agree on is recorded as user. It’s the Socratic path: the AI facilitates, you decide.

Use it when you’re designing something new or reworking existing architecture — “design”, “architect”, “let’s think about…”. For implementing within decisions that already exist, you don’t need this; the agent just pulls the brief and builds.

Starting one

Ask your agent, in plain language:

Let’s design the rate limiter. Walk me through it with Trurlic.

The agent calls advance(component="rate-limiter", task_type="new_component", mode="interactive") and follows the loop. In interactive mode, comprehension gates are on — the agent can’t record a decision until you’ve engaged with it.

How it flows

advance(component, task_type, mode="interactive")
        │
        ▼
   gated step → requires_user_input: true
        │        agent presents the step's prompt — a real question,
        │        not a quiz — and waits for your answer
        ▼
   you respond → agent records the decision (attribution: user)
        │        and calls advance again with step_evidence
        ▼
   … repeat through the task's steps …
        ▼
   ready: true → agent pulls the brief and implements

Each gated step needs a substantive response (at least 20 bytes) before the workflow advances. See Workflow Engine for the step sequences and gating.

What gets recorded

As you agree on choices, the agent calls record_decision with attribution="user" — the choice, your reasoning, rejected alternatives, and the code the decision touches. Because you were in the loop, these land human-authored: no ⚠ flag, no review debt. When several related decisions share an approach, the agent may synthesize a pattern.

Interactive vs agent mode

	Interactive	Agent
Who decides	You and the AI, together	The AI, from the source
Comprehension gates	On	Off
Attribution	user	agent (flagged ⚠)
Best for	design, architecture, learning	implement/fix, bootstrapping

learn tasks are always interactive; bootstrap is always agent. See Core Concepts for the full comparison and Bootstrap for the autonomous path.