DeepSeek’s team-agent prototype puts authorization in the host
DeepSeek’s public super-agent prototype turns authorized GitHub or Feishu requests into separate work items, with host-enforced permissions and an explicit warning that it is research code.
DeepSeek’s Harness team committed its public super-agent prototype on October 31.
It listens to one GitHub repository and optionally Feishu, then creates a separate agent session and directory for an admitted work item. It can answer back in the originating conversation, including with a pull request, while a panel collects matters needing human attention. The repository says it is a research prototype, differs from the internal implementation and does not implement every direction discussed in the associated research. [1][3]
The important design choice is where authority lives. The host enforces allowed people, written rules and the scope assigned to each work item. A comment or quoted document is input to a decision, not permission to rewrite those boundaries. Messages from subscribed threads can supply context without granting additional authority. [2]
Sources & details 3 sources
Discuss in your voice app
Take the reporting, visual explanations and original sources into a conversation. Copy the brief, paste it into your preferred assistant, then start its voice mode.
Paste the brief into your voice app to discuss this reporting.Reporting sources
- DeepSeek — prototype README
Public behavior, single-repository scope and explicit research-only warning.
- DeepSeek — authorization design
Host-enforced allowed lists, rules, mandates and context-versus-authority distinction.
- DeepSeek — commit history
October 31 initial commit; event date, not an independently established commercial release.
Images & figures
DeepSeek’s prototype admits work through host-enforced authorization and separates each work item’s session and directory. Later conversation can add context without enlarging its authority.
Euangelion · Source · Usage terms