Skip to content
Proposals/Launch snapshot-isolated child agents from hoste
proposalteamP5Worth a lookQCQwen Code

Launch snapshot-isolated child agents from hosted Qwen Code parents

Qwen Code v0.25.3 lets hosted parents spawn snapshot-isolated children that read a frozen workspace without mutating files. Pair that with pinned Skills/MCP/Hooks on AgentDefinitions, Plan mode, and Shell Full Access switching so exploration stays isolated while sessions keep history.

Why this loop

Qwen Code v0.25.3 (#14265) lets hosted parent agents launch snapshot-isolated child agents that read a frozen workspace copy without changing files, running commands, or affecting the parent—use that for parallel investigation instead of mutating the parent session. Managed sessions from published AgentDefinitions can load static Skills and pass pinned MCP server and Hook catalog selections to Hosted Runtimes (#14319), and the composer now connects to those pinned Skills (#14602). Enable Plan mode with inline attachments (#14531) and put sessions on the Web Shell approval mode allowlist (#14345) before switching Workspace-bound qwen-code Sessions among default, auto-edit, yolo, and Shell Full Access; that switch keeps history, Workspace, and drafts (#14323). Closing a parent now stops each child's running Turn (#14218). Eligible legacy cancellations require confirmation through the daemon or WebShell Continue flow, while explicitly user-cancelled work stays blocked (#14087). Do not assume anything from the 18 omitted changelog entries.

Proposed actions

  1. Upgrade to Qwen Code v0.25.3, then from a hosted parent agent launch a snapshot-isolated child agent against the frozen workspace copy; instruct it to inspect only and never change files, run commands, or write back to the parent.
  2. Publish an AgentDefinition that loads static Skills and pins the MCP server and Hook catalog selections so new managed sessions pass those pins to Hosted Runtimes.
  3. In the Web Shell composer, turn on Plan mode, attach files inline, and send the next turn using the pinned Skills already connected to the composer.
  4. Add this Workspace-bound qwen-code Session to the Web Shell approval mode allowlist, then switch it in place among default, auto-edit, yolo, and Shell Full Access so history, Workspace, and drafts are not dropped.
  5. Close the parent session when the investigation is done so each child session's running Turn stops; confirm any eligible legacy cancellation through the daemon or WebShell Continue flow, and leave explicitly user-cancelled work blocked.

Agent prompt

Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI cost

Copilot Chat / Agent mode

DevAgentRadar → GitHub Copilot

You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.

Context

Assistant: Qwen Code Proposal: Launch snapshot-isolated child agents from hosted Qwen Code parents Summary: Qwen Code v0.25.3 lets hosted parents spawn snapshot-isolated children that read a frozen workspace without mutating files. Pair that with pinned Skills/MCP/Hooks on AgentDefinitions, Plan mode, and Shell Full Access switching so exploration stays isolated while sessions keep history. Primary source: https://github.com/QwenLM/qwen-code/releases/tag/v0.25.3

Why it matters

Qwen Code v0.25.3 (#14265) lets hosted parent agents launch snapshot-isolated child agents that read a frozen workspace copy without changing files, running commands, or affecting the parent—use that for parallel investigation instead of mutating the parent session. Managed sessions from published AgentDefinitions can load static Skills and pass pinned MCP server and Hook catalog selections to Hosted Runtimes (#14319), and the composer now connects to those pinned Skills (#14602). Enable Plan mode with inline attachments (#14531) and put sessions on the Web Shell approval mode allowlist (#14345) before switching Workspace-bound qwen-code Sessions among default, auto-edit, yolo, and Shell Full Access; that switch keeps history, Workspace, and drafts (#14323). Closing a parent now stops each child's running Turn (#14218). Eligible legacy cancellations require confirmation through the daemon or WebShell Continue flow, while explicitly user-cancelled work stays blocked (#14087). Do not assume anything from the 18 omitted changelog entries.

Suggested actions

  1. Upgrade to Qwen Code v0.25.3, then from a hosted parent agent launch a snapshot-isolated child agent against the frozen workspace copy; instruct it to inspect only and never change files, run commands, or write back to the parent.
  2. Publish an AgentDefinition that loads static Skills and pins the MCP server and Hook catalog selections so new managed sessions pass those pins to Hosted Runtimes.
  3. In the Web Shell composer, turn on Plan mode, attach files inline, and send the next turn using the pinned Skills already connected to the composer.
  4. Add this Workspace-bound qwen-code Session to the Web Shell approval mode allowlist, then switch it in place among default, auto-edit, yolo, and Shell Full Access so history, Workspace, and drafts are not dropped.
  5. Close the parent session when the investigation is done so each child session's running Turn stops; confirm any eligible legacy cancellation through the daemon or WebShell Continue flow, and leave explicitly user-cancelled work blocked.

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/qwen-code-v0-25-3-launch-snapshot-isolated-child-agents-from-hosted-qwen and mark Applied, Skipped, or Failed. Proposal id: cd37e3f6-7579-4d5a-8952-34f49a82b4f1

Your job

  1. Restate the change in one sentence.
  2. Propose a minimal plan for my repo (or a throwaway pilot).
  3. Implement only what I approve; prefer small diffs and tests.
  4. Call out risks (permissions, breaking APIs, cost).

Start by confirming you understood the proposal.

agentmcpmodelsecurityideRelease source ↗

Your loop

This browser · no sign-in · not shared as “you”

After you run the prompt

Only you can mark this. Agents cannot write your loop.

Your decision stays on this device. A public tally appears after a few votes.

Originating release signal

QCQwen Codev0.25.3Oct 19, 2026

v0.25.3 Release v0.25.3

See the complete change list below. · No known breaking changes. · Adds optional model-generated layouts to Test Failure Explorer, with validation and fallback to the fixed view (#14269) · The Web Shell mobile composer no longer shows the ↑ and ↓ history buttons, reducing clutter while keeping history search available. (#14220) · +66 more changes
Verified excerpt — the source's own words
<!-- qwen-release-notes:v1 -->

Highlights

See the complete change list below.

Breaking Changes

No known breaking changes.

Complete Change List

Features

  • Adds optional model-generated layouts to Test Failure Explorer, with validation and fallback to the fixed view (#14269) by @qqqys

  • The Web Shell mobile composer no longer shows the ↑ and ↓ history buttons, reducing clutter while keeping history search available. (#14220) by @pomelo-nwu

  • The Web Shell trajectory panel now includes a collapsible Recent executions section that shows live requests and tools with elapsed time and final durations. (#14071) by @qqqys

  • Managed sessions created from published AgentDefinitions can now load static Skills and pass pinned MCP server and Hook catalog selections to Hosted Runtimes. (#14319) by @doudouOUC

Excerpt ends here — this release continues at the source ↗.

Primary source ↗