Feature request for product/service
Cloud Agent (GitHub, Slack, Linear,…)
Describe the request
We use Linear Loops to prepare implementation plans before delegating issues to Cursor. The workflow is:
A loop (Linear Agent) analyzes a triage ticket with Code Intelligence and posts a detailed implementation plan as a comment: root cause, files to modify, proposed steps, tests, open questions.
The team reviews the plan and answers the open questions in the thread.
We delegate the issue to Cursor to implement the validated plan.
The problem: when the Cloud Agent session starts, the context payload injected into the agent contains the issue description, the project description, and human-authored comments, but every comment authored by the Linear Agent app user is silently dropped. In our case, a thread where the parent comment is the plan (authored by “Linear”) and the reply is a human answer shows up in the payload with only the human reply. The plan itself, which is the most important piece of context, never reaches the agent.
The result is that the agent implements from the raw ticket description plus fragments of human replies, ignoring the reviewed plan (file paths, constraints, test requirements) that the team explicitly validated for it.
Steps to reproduce
Create a Linear Loop that posts a comment on an issue (any content).
Have a human reply to that comment.
Delegate the issue to Cursor.
Inspect the agent session’s injected context: the human reply is present under other_comment_threads, the agent-authored parent comment is absent.
Why this matters
Linear is pushing Loops as a first-class automation pattern, and “an agent prepares, a human validates, Cursor implements” is a natural pipeline on top of it. Filtering agent-authored comments (presumably to avoid agent-to-agent prompt injection) breaks that pipeline with no way to opt out.
Suggested solutions (any of these would work)
A team-level setting in the Cursor dashboard to include comments from specific Linear app/agent authors (allowlist).
Honoring Linear’s onBehalfOf field: treat agent comments made on behalf of a human as human-authored.
A convention the integration respects, e.g. include agent comments when the delegating user reacts to them or when they are in the same thread as a human reply.
Current workaround
We now have the loop write the plan into the issue description instead of a comment, since the description is always injected. It works, but it mixes generated content into the ticket body and loses the natural review-in-thread flow.