mirror of
https://github.com/deepseek-ai/deepseek-harness
synced 2026-08-15 21:04:50 +00:00
91247babd9
A recorded scenario where the model writes a workspace file in one write call with sandbox_permissions=danger-full-access + justification, the scripted client allows once, and the escalated write lands — the fs twin of escalation-approved, covering the fs escalation approval arc (only bash had a snapshot before). Uses a new workspace-internal file to avoid the read-before-edit detour, keeping the recorded path single and deterministic.
12 lines
721 B
JSON
12 lines
721 B
JSON
{
|
|
"steps": [
|
|
{ "op": "initialize" },
|
|
{ "op": "newSession" },
|
|
{ "op": "setConfigOption", "configId": "permission", "value": "workspace-write" },
|
|
{ "op": "prompt", "text": "Use the write tool (NOT bash) to create escalated.md in the current directory containing exactly the single line: escalated. An equivalent write was denied earlier, so make this one single write call with sandbox_permissions set to danger-full-access and the justification 'the user asked to escalate this write'. Do not call write without sandbox_permissions first. I will approve the permission prompt. After the result, reply with exactly the single word DONE." }
|
|
],
|
|
"permissionAnswers": [
|
|
{ "kind": "allow_once" }
|
|
]
|
|
}
|