DeepSeek Harness RC.8 Brings Native Multimodal Input and Pluggable Claude Code/Codex Subagents
Around August 20, DeepSeek's open-source agent framework Harness released v0.1.0-rc.8, focusing on two fronts: filling in multimodal input gaps and integrating external coding tools into a modular subagent system. Multiple authors (@teortaxesTex, @aigclink, @机器之心, @智东西, @APPSO) reported the same changes, while @lxfater noted the release closes the framework's previous multimodal shortcoming.
Confirmed
- Multimodal: The DeepSeek model adapter can enable native image requests; core commands such as /goal and /plan accept mixed text-and-image input, and the @ menu can reference local files and past sessions.
- Modular setup: Claude Code and Codex can be installed on demand as standalone Profile Bundles plugged into the Harness subagent system.
- Codex improvements: Non-interactive permission mode and multiple named instances, enabling parallel Codex usage.
Why it matters
- Image and file/session reference support removes a previously noted weakness, making Harness more practical for mixed text-and-image tasks.
- Bringing third-party coding agents under one subagent umbrella reflects Harness's modular orchestration direction and lowers the barrier to combining different coding agents.
2026-08-20 ~ 2026-08-21 · 8 related posts
Primary sources
- DeepSeekHarness RC.8 adds ClaudeCode/Codex sub-agents and multimodal support — AI寒武纪 · 2026-08-20
- [source] DeepSeek Harness Update: Multi-Codex Instances & Multimodal Support — teortaxesTex · 2026-08-20
- [source] DeepSeek Harness RC.8 adds native multimodal and Claude sub-agent orchestration — aigclink · 2026-08-20
- DeepSeek Harness update enhances multimodal capabilities — lxfater · 2026-08-21