This feature request is specifically for the Cloud Agent + Linear integration.
The Problem
When the Cursor agent picks up a Linear ticket and starts working on it, it behaves like a junior developer who jumps straight into coding without gathering requirements first. Specifically:
It reads the ticket title/description and immediately starts making code changes
It does NOT ask clarifying questions upfront (e.g. “Where should this be implemented?” or “Which approach do you prefer?”)
It does NOT check relevant files first to understand context before asking
The result is wasted work, wrong assumptions, and code that often has to be thrown away
Proposed Solution: 2-Stage Clarification Flow
Stage 1 — Upfront question based on the prompt:
Before touching any files, the agent should analyze the Linear ticket and ask 1-3 targeted clarifying questions about scope, approach, and intent. This mirrors how a senior engineer would first ask “what exactly are we trying to solve here?”
Stage 2 — Post-file-scan clarification (when needed):
After the agent scans the relevant files and understands the codebase, it should pause again if it finds ambiguity (e.g. two possible implementation locations, conflicting patterns, unclear ownership) and ask a second round of targeted questions BEFORE writing any code.
Only after both stages are satisfied should it proceed to implement.
Why This Matters
The current behavior causes agents to produce half-baked PRs that require significant human rework. The 2-stage approach would make the Linear integration production-ready and actually useful for teams. Senior engineers always clarify requirements before coding — the agent should do the same.
Hey, thanks for the detailed feature request. Good news, part of what you’re describing already works via Plan Mode.
In Plan Mode, the agent first analyzes the task and asks follow-up questions using a structured Ask Question picker before it starts writing code. That matches your Stage 1. After it scans relevant files, it can ask another round of questions if something is ambiguous. This flow is also available in cloud agents. You can enable it with Shift+Tab or by typing /plan.
What’s not available out of the box yet is having a Linear-triggered automation agent default to clarify-before-code mode. Right now it runs autonomously and starts coding right away. Plan Mode with follow-up questions already works in cloud agents in the Agents Window, but Linear automations don’t enter it by default. I’ve noted your scenario as a signal.
I’ll keep the thread open. If it gets more votes and replies, that’ll help with prioritization. Let me know if Plan Mode helps cover your use case for now.
Good question. The key thing is where you’re starting the cloud agent from.
Plan Mode is available for cloud agents if you start them from the Agents Window inside Cursor, not from the web UI. Here’s the flow:
Open the Agents Window in Cursor
Switch to Plan Mode via Shift+Tab or type /plan
When starting the agent, pick a cloud remote runtime
In this setup, the agent will first explore the task and ask clarifying questions via the Ask Question picker, and only then start writing code, same as local mode.
What’s not there yet: Plan Mode isn’t exposed in the web UI at cursor.com/agents, so you won’t see it there. Also, Linear triggers for auto agents don’t enter Plan Mode by default yet, which is what you’re asking for in the request.
If you’re using the API or SDK, you can explicitly pass mode: "plan" when creating a cloud agent.
Let me know if starting it via the Agents Window works for your case.