Adding Claude Like hooks to Qwen Code

[rinadelph]

What would you like to be added?

I want to be able to control what happens before and after each tool call and even being able to inject status infromation to the agents.

Why is this needed?

This creates more auditable and deterministic agents for multi agent collaborative workflows.

Additional context

No response

[MJRT]

any update?

[ppiankov]

+1 — here are concrete use cases from running hooks across Claude Code and Cline today:

Use case 1: Enforce MCP tools for files with secrets

We run a local MCP server (pastewatch) that provides pastewatch_read_file / pastewatch_write_file — these replace secrets with PW{TYPE_N} placeholders so the API never sees raw credentials.

A PreToolUse hook on Read|Write|Edit scans the target file for secrets. If found, the hook blocks the native tool and tells the agent to use the MCP tool instead. Structural enforcement — the agent cannot bypass it.

Without hooks, we can only put instructions in QWEN.md saying please use pastewatch tools for secrets. The agent may ignore this, especially with weaker models, and send raw API keys to the cloud.

We currently recommend Cline over Qwen Code for sensitive data work specifically because of this gap.

Use case 2: Block dangerous bash commands

A PreToolUse hook on Bash that blocks destructive operations (recursive deletes on root, force pushes, hard resets, dev servers that lock the session, etc). No way to structurally prevent these in Qwen Code today.

Use case 3: Auto-format after edits

A PostToolUse hook on Write|Edit that runs gofmt, black, prettier after the agent modifies a file.

Use case 4: Stop hook for verification

A Stop hook that checks whether tests were run before the agent finishes. If not, it re-injects a prompt telling the agent to run tests.

---

Suggested protocol (same as Claude Code + Cline)

Configuration in settings.json:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [{ "type": "command", "command": "~/.qwen/hooks/bash-guard.sh" }]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [{ "type": "command", "command": "~/.qwen/hooks/auto-format.sh" }]
      }
    ]
  }
}

Protocol:

• Hook receives JSON on stdin: {"tool_name": "...", "tool_input": {...}}
• PreToolUse: exit 0 = allow, exit 2 = block. stdout = message back to agent.
• PostToolUse: exit code ignored. For side effects only.
• matcher supports pipe-separated tool names: "Read|Write|Edit"

Adopting this protocol would make existing Claude Code / Cline hooks work in Qwen Code without modification. OpenCode has a similar request in anomalyco/opencode#12472. Standardizing across agents benefits everyone.

[Mingholy]

#1988 has submitted a basic implementation of hookos
More hooks are on the way. #2033