[Feature Request] Support for Linear Sub-issue Trees and Unified Workspaces

[bravous]

Context
Currently, the agent handles individual issues in isolation. However, in platforms like Linear, complex tasks are often broken down into a Parent Story with multiple Sub-issues (tasks). To solve these effectively, the agent needs the full context of the parent and all related children to ensure architectural consistency.

Proposed Feature
Enable the agent to ingest a Linear Parent ID, fetch the entire tree of sub-issues, and treat them as a single "Mission" or "Worktree."

Desired Workflow

1. Context Fetching: When provided with a Parent ID, the agent should query the Linear API for all sub-issues (e.g., US-1469 through US-1473 in the attached example).
2. Unified Workspace: Create one branch/worktree for the entire story. The agent should solve the sub-issues sequentially or holistically within this single environment.
3. Granular Updates: As the agent completes the logic for a specific sub-issue, it should post a progress comment/status update to that specific sub-issue ID in Linear, while keeping the main PR linked to the Parent Story.
4. Final Submission: Once all sub-tasks in the tree are addressed, the agent submits the unified PR for review.

Trigger Mechanism (Ideas)
We could implement this in two ways:

• Explicit Label: A specific label (e.g., cyrus-solve-tree) applied to the parent issue triggers the recursive fetch.
• Agent Intelligence: The agent automatically checks if an issue has sub-issues. If it does, it makes an "educated guess" or asks for confirmation: "I see this issue has 5 sub-tasks. Should I solve the entire tree in one branch?"

Use Case Example
As seen in the attached screenshot (API Versioning Deprecation):

• Parent: API Versioning Deprecation and Changelog
• Sub-tasks: Build interceptor, Create config, Set up changelog, Add docs, Add tests.
• Value: Solving these individually would be difficult because the "interceptor" depends on the "config" and the "tests" depend on both. A unified worktree allows the agent to build the config first, then the interceptor, then the tests, ensuring a passing build.

How to use this:

1. Go to the GitHub repository for Cyrus.
2. Click New Issue -> Feature Request.
3. Paste the content above.
4. Upload your screenshot to the issue so they can see the Linear hierarchy example.

Would you like me to refine the technical "Trigger Mechanism" section to suggest a specific API flow for how the agent should detect these sub-issues?

Enabling Spec-Driven Development & "Bite-Sized" Review
Beyond simple task automation, this feature enables a true Spec-Driven Development (SDD) workflow. By allowing a human (or an AI like Claude) to pre-plan a complex story into granular sub-tasks—including small manual additions like "Update README"—we create a clear roadmap for the agent.

This approach solves the "Eating the Elephant" problem in two ways:

1. For the Agent: Operating within a single worktree for the entire sub-issue tree ensures architectural consistency across related tasks, preventing merge conflicts and "context drift."
2. For the Team: By commenting on individual sub-issues as it progresses, the agent provides "byte-sized" updates. This allows the team to review and course-correct small pieces of logic in real-time, preventing the "review fatigue" typically caused by massive, multi-scope PRs.

Essentially, this transforms Cyrus from a "single-task bot" into a background collaborator capable of executing a coordinated plan with senior-level communication.

[Connoropolous]

try with the 'orchestrator' label :)

[bravous]

Ah maybe I misread the manual, I was in the understanding that the orchestrator label would CREATE sub issues and then start working on it.

When I ask a simple question Cyrus is able to understand it is a question by analyzing the issue, no label needed. Maybe her could do the same if he by default pulled or checked sub issues and then determine if it is a story>task set.

Another though I had was that I could provide a specify skill or maybe multiple in that scenario I can create powerful flows like

Take the issue, look for sub issues, run tests, create screenshot of final result and post as comment to the issue.