Agent Atlas / Coding agents

QwenLM/qwen-code

A terminal coding agent that now documents IDE, desktop, daemon, SDK, messaging, multi-provider, memory, skills, sandbox, and agent-team surfaces.

Coding agent CLI Full dossier Project documented Reviewed August 26, 2026

Intended operator

Developers using an interactive terminal, headless automation, an IDE, or a shared agent service.

Primary surfaces

terminal and headless CLI / IDE and desktop clients / daemon, SDK, and messaging clients

Tracked repository

QwenLM/qwen-code

Teaching mechanism

One daemon can expose a shared agent session over ACP

The documented daemon mode serves one agent session to multiple clients over HTTP and server-sent events, turning client identity and session ownership into protocol questions.

Evidence limitThe daemon is documented as experimental; multi-client reliability and authority require source and interoperability testing.

Inspect the source

System Map

Eight inspection boundaries

Model boundary

OpenAI, Anthropic, Gemini, Qwen, compatible, and local provider paths can be switched at runtime.

Context and memory

Auto-memory, skills, hooks, session management, and worktrees shape durable task context.

Tools and execution

Terminal, IDE, MCP, LSP, computer-use, and headless surfaces expose several execution boundaries.

Permissions and isolation

The project documents sandbox and auto modes; exact enforcement across surfaces remains to be traced.

Sessions and recovery

Interactive, headless, daemon, and SDK clients share session-management concerns.

Delegation

Subagents, agent teams, and delegation through ACP are documented capabilities.

Human control

Plan mode, interactive commands, IDE clients, and an agent arena expose several supervision shapes.

Evidence surface

The public repository, docs, releases, tests, and protocol surfaces provide the starting evidence.

Learn Through This Project

10 lessons

Chapter 1 Before the First Model Request Chapter 2 The Invisible Prompt Chapter 3 Tool Calls: Contract, Authorization, Execution Chapter 4 Loops, Workflows, and Recovery Chapter 5 Delegation Is a Distributed System Chapter 6 Memory Is a Typed Storage Contract Chapter 7 The Approval Trap Chapter 8 The Human Control Loop Chapter 9 Trust Must Make Its Case Chapter 10 The Future Has Three Branches

Compare Its Boundaries

8 labs

Comparison labTool Contracts Comparison labGoverned Memory Comparison labDelegation Receipts Comparison labApproval Boundaries Comparison labHuman Workbench Comparison labSessions And Recovery Comparison labEvidence And Evaluation Comparison labMaturity Signals

Latest Reporting

8 articles

Qwen Code Draws a New Sandbox Boundary. Saving the Policy Still Has a Gap.

September 23, 2026 — Linux tool confinement now leaves the agent’s model connection on the host. The public review shows why the settings file that selects that boundary deserves just as much scrutiny. QwenLM/qwen-code

Qwen Code Stops Calling a Rerun a Resume

September 19, 2026 — Two September 18 changes separate a recoverable interruption from a fresh start: refuse a missing journal, then make the interrupted run visible. Neither promises automatic recovery. QwenLM/qwen-code

API Discovery: Give the Reviewer Fewer Tools Than the Builder

September 15, 2026 — Qwen Code’s workflow API can set one agent’s reasoning effort and remove tools from that step. Those choices also change which saved result a resumed workflow may reuse. QwenLM/qwen-code

Builder Work: A Socket Path Is Not a Permission Boundary

September 2, 2026 — Qwen Code’s Claude-assisted inbox hardening shows the reusable craft: treat an endpoint and its credential as one capability, then erase both before a child process can inherit them. QwenLM/qwen-code

Builder Work: A Safety Check Can’t Read a Blank Form

August 31, 2026 — Qwen Code’s AUTO-mode repair turns an empty MCP permission prompt into a bounded, visibly incomplete policy view—a small interface with a serious security lesson. QwenLM/qwen-code

Builder Work: The Model Belongs to the Session, Not the Settings File

August 23, 2026 — Qwen Code’s agent-assisted daemon patch turns a fragile global preference into a tested per-session record—and offers a durable-state lesson other builders can steal. QwenLM/qwen-code

Qwen Code Saves the Shape of a Finished Workflow

August 21, 2026 — Qwen Code is teaching a finished multi-agent run to leave behind a reviewable shape—rather than a pile of logs that dies with the terminal. QwenLM/qwen-code

Coding Agents Are Building Themselves—But Git Can Prove Only Part of It

August 11, 2026 — Across 40,345 unique commits in ten public agent codebases, coding-agent credit becomes sharply more visible in 2026—and the missing receipts matter as much as the names that appear. openai/codex, google-gemini/gemini-cli, anomalyco/opencode, QwenLM/qwen-code, cline/cline, aaif-goose/goose, OpenHands/OpenHands, charmbracelet/crush, mistralai/mistral-vibe, Kilo-Org/kilocode

Primary Sources

Qwen Code repository

Primary public source for the terminal agent and its expanding client surface.

Open Questions

How are authority and session ownership kept coherent when one agent serves several clients?

Evidence state describes what this Atlas profile can support today. It is not a product score and it is not evidence that uninspected capabilities are absent.