Queries Sent After Switching from Act to Plan Mode Are Flagged by Azure “jailbreak” Filter and Return 400 Error

Plugin Type

VSCode Extension

Cline Version

v3.53.1

What happened?

Toggling between Plan and Act modes injects text that AzureOpenAI’s content filter always misclassifies as a jailbreak policy violation, resulting in a 400 error.

Steps to reproduce

1. Start a conversation using Plan mode.
2. Switch the same session to Act mode and send a request to an OpenAI model via an OpenAI-compatible API.
3. Observe an HTTP 400 response.
4. Alternative: Start in Act mode, switch to Plan mode, then send a request — also results in HTTP 400.

Response:

"modelId": "azure.gpt-5.1",
    "providerId": "openai",
        "type": null,
        "param": null,
        "code": "400",
        "provider_specific_fields": {
            "innererror": {
                "code": "ResponsibleAIPolicyViolation",
                "content_filter_result": {
                    "hate": {
                        "filtered": false,
                        "severity": "safe"
                    },
                    "jailbreak": {
                        "filtered": true,
                        "detected": true
                    },
                    "self_harm": {
                        "filtered": false,
                        "severity": "safe"
                    },
                    "sexual": {
                        "filtered": false,
                        "severity": "safe"
                    },
                    "violence": {
                        "filtered": false,
                        "severity": "safe"
                    }
                }
            }
        }

Provider/Model

Azure OpenAI gpt-5.1

System Information

• macOS Tahoe 26.2
• VS Code 1.108.2

Further Information

The problem was addressed in this issue for a previous version of Cline (probably since v3.12). However, it still persists in current version