[FEATURE] Let the agent programmatically create and manage conversation threads

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I regularly run several independent streams of work and the agent itself can't help me organize them, thread management is entirely manual on my side.

Today the in-chat agent can spawn in-session subagents (ephemeral; only the final message comes back), and I can manage sessions myself via /bg, Agent View, /resume, and /branch. But the agent has no way to create or manage top-level conversation threads on my behalf.

So when a conversation grows and I want to split off a separate, independent line of work, or have the agent kick one off, check on it later, and tell me how it went, I have to stop, leave the conversation, manually open/resume/rename/archive sessions in the sidebar, and re-establish context each time. The agent can see the work that should be parallelized but can't act on it. That manual juggling is the bottleneck, not the work itself.

Proposed Solution

Expose agent-callable operations for managing top-level threads (not just subagents), so the agent can orchestrate them from within a conversation:

• Create a new thread (project-scoped or projectless), optionally seeded with a task prompt
• Dispatch several independent threads in parallel
• Continue an existing thread with a follow-up message
• Inspect a thread's status, summary, and latest output
• List / search recent threads
• Rename, archive/unarchive (pin/unpin nice-to-have)

In the desktop app, these actions should reflect live in the session sidebar so I can see what the agent created and jump in. The plumbing largely exists already at the SDK/API layer (Agent SDK resume: sessionId, Managed Agents API with server-side persisted history) — this request is about surfacing it as tools the conversational agent can call.

OpenAI's Codex already exposes this to its agent: from inside a conversation it can create new threads (project or projectless), dispatch parallel threads, continue/inspect/list/search, and rename/pin/archive them. That's the experience I'm asking for in Claude Code.

Alternative Solutions

_No response_

Priority

Low - Nice to have

Feature Category

Developer tools/SDK

Use Case Example

Example scenario:

1. I'm in a Claude Code conversation working on a bug fix in my main repo.
2. Mid-conversation I realize there are two unrelated follow-ups: updating docs in the same repo, and investigating a flaky test in a sibling repo.
3. With this feature, I ask the agent to "spin these off into their own threads and keep working." It creates two new threads (one project-scoped, one in the sibling repo), seeds each with the task, and keeps my current conversation focused on the bug fix.
4. Later I ask "how are those threads doing?" the agent lists them with status and a one-line summary of each, and I tell it to continue one with a follow-up and archive the other.
5. This saves time because I never leave my main conversation or manually open/resume/rename sessions in the sidebar — the agent handles the orchestration and I stay in flow.

Additional Context

Reference for the equivalent capability inOpenAI Codex agent thread management: create (project/projectless), dispatch parallel, continue, inspect, list/search, rename, pin/unpin, archive/unarchive.

Related issues in this repo:

• #38175 — Multiple independent chat threads (tab/sidebar UI) — the UI side of the same need
• #55421 — Expose session ID to the agent during execution (closed; partial building block)