Skip to content
Productivity
Command

/launch-new-session

Spin off a new AI session for unrelated work, a fresh continuation, or a worker

From plugin
nimbalyst
2.1k54 skills3 agents54 commands3 MCP
> /plugin marketplace add nimbalyst/nimbalyst
> /plugin install nimbalyst-wiki@nimbalyst

Command definition

launch-new-session.md
description: Spin off a new AI session for unrelated work, a fresh continuation, or a worker

/launch-new-session Command

Spin off a new AI session, e.g. "/launch-new-session and keep going on the tests" or "/launch-new-session to fix that unrelated bug". Pick the mode from what the user asked for:

  • **Independent** (`isolated: true`): for work unrelated to the current task, e.g. "fix that other bug", "chase these down separately", "commit it on its own". The new session is a top-level session that shares no files-edited, tabs, or workstream overview with this one. Fire and forget: do not wait for it or check on it. `spawn_session` rejects `isolated: true` with any `reporting` other than `"none"`; if you need the result back, it is a Worker.
  • **Continuation** (`besideCaller: true`): for the same work in a fresh session, usually to escape a long context. The new session sits beside this one in the session list, so a chain of handoffs stays flat.
  • **Worker** (neither flag): for part of a task this session is coordinating, when the user asked it to orchestrate or parallelize work. The new session nests under this one in the session tree. Use `reporting: "final"` so the result comes back.

Reporting defaults to `"none"`: the calling session is not told when the new session finishes. Use `"final"` only for workers, or when the user asks for the result to come back here.

Steps

When the user types `/launch-new-session [task description]`:

1. **Construct a self-contained handoff brief.** The new session will not see this conversation, so the brief must stand on its own. Include:

  • The task in 1-2 sentences (what to do, what success looks like)
  • Relevant file paths the new session should look at
  • Constraints or decisions already made in this session that affect the work
  • This pointer at the bottom: "For more context on what led to this task, call `get_session_summary` with `sessionId=<the current session id>`."

2. **Pick the mode** (see above). Unrelated work is Independent; "keep going", "continue", or a long context is Continuation; Worker only when this session is coordinating the work. When unsure between Independent and Continuation, ask whether the new work belongs to the current task.

3. **Decide on `useWorktree`.** Default to `false`. The default already inherits the caller's working directory: if the current session is running in a worktree, the spawned session runs in that same worktree (its edits land where the user is looking). Only set `true` when the user's phrasing implies the new session needs its OWN new branch / working directory — e.g. "in a new worktree", "in parallel without conflicts", "without touching my current branch". Note: `isolated` and `useWorktree` are independent — isolated alone still inherits the caller's working directory but separates the session record; combine with `useWorktree: true` to also branch off into a fresh worktree.

4. **Decide on `reporting`.** Omit it (fire and forget) for Independent and Continuation. Use `"final"` for a Worker, or when the user wants the result back in this session ("...and tell me when it's done").

5. **Decide on model.** By default the new session uses the app's global default model. Override only when the user asks for it:

  • If the user names a model ("...with opus", "...using sonnet"), pass that as `model` (e.g. `model: "claude-code:opus"`).
  • If the user says "same model", "keep the current model", or similar, pass `inheritModel: true` so the new session uses the caller's model.
  • `model` wins over `inheritModel` when both are set.

6. **Call `spawn_session`** with:

  • `prompt`: the handoff brief from step 1
  • `title`: a short descriptive title (e.g. "Finish auth tests")
  • `isolated: true` (Independent) or `besideCaller: true` (Continuation), per step 2
  • `useWorktree`: per step 3
  • `reporting`: per step 4
  • `model` / `inheritModel`: per step 5 (omit both to use the global default)

7. **Report back to the user** with:

  • The new session id
  • A one-line summary of what was handed off
  • The mode used: independent, continuation, or worker (and the worktree, if `useWorktree` was set)

Notes

  • Do NOT pre-summarize the parent session in the prompt beyond what's needed to act on the task. The new session can call `get_session_summary` if it needs more.
  • Do NOT spawn a new session for trivial follow-ups that the current session can handle directly — `/launch-new-session` is for parallel work or context-escape hand-offs.
  • The new session inherits the workspace; cross-workspace spawning is not supported.
Read more
@nimbalyst
Ships withnimbalyst

Nimbalyst - The open-source visual workspace for Claude Code, Codex, and OpenCode. Run multiple coding agents in parallel, edit their work visually in markdown, mockups, and diagrams, and track tasks. Free, MIT-licensed desktop app for macOS, Windows, Linux, with mobile companion for iOS and Android.

2.1k16 skills3 agents3 MCP
Get the whole plugin

Stars on GitHub

Momentum

stars in total
2.1k
Star history of nimbalyst/nimbalyst on GitHub, a chart by star-history.com

Other commands on nimbalyst.

analyze-sessions

analyze-sessions

Command

Audit recent AI coding sessions to find repeated mistakes, speed losses, and missed Nimbalyst tool usage — then propose harness…

@nimbalyst@nimbalystInstall Command
audit-updates

audit-updates

Command

Triage pnpm audit findings and produce a prioritized, supply-chain-cautious package-update plan, then apply approved batches

@nimbalyst@nimbalystInstall Command
autofix-issues

autofix-issues

Command

Survey recently triaged GitHub issues, propose the ones safe to fix without a product decision, and fan the selected ones out to…

@nimbalyst@nimbalystInstall Command

More commands from other plugins.

analyze-cohorts

analyze-cohorts

Command

Perform cohort analysis on user data — retention curves, feature adoption, and engagement trends

@phuryn@phurynInstall Command
again

again

Command

PUA Again — 用于再试试、换个方法、别放弃、为什么还不行、同一路径失败多次;要求停止微调同一思路,改用本质不同方案。

pua2mo20k
@tanweai@tanweaiInstall Command