DeepSeekHarness dev analysis: high efficiency, Koishi-like plugins
歸藏的AI工具箱 · wechat · 2026-08-14
An analysis of the DeepSeekHarness repository reveals:
- Plugin System: Highly similar to Koishi, suggesting architectural reuse.
- AI-Assisted Dev: Significant contribution from AI coding tools (likely Claude Code).
- Efficiency: 840k lines of code and 10k+ commits in 65 days.
- Architecture Shift: Transitioned from TUI to WebUI; test-to-production code ratio is 1:1.
- Ecosystem: Reached 80k+ stars in 20 hours; the plugin ecosystem is growing fast with some low-quality spam.
- Academic Support: An 88-page paper was released focusing on plugin hot-plugging and system stability.
Related event: DeepSeek Harness: ~20% of Commits Were Made by Codex(3 posts)→
More from Companies & People
- Wildeford: OpenAI repeating past behavior, opposite of what AI safety needs — peterwildeford · 2026-10-09
- Legal Group Psst Is Representing the Three Fired OpenAI Employees Pro Bono — GarrisonLovely · 2026-10-09
- Bostrom, State of AI 2026 and Reflection's 501B Beam Headline Stacked MTS Livestream — nathanbenaich · 2026-10-09
- Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect — TechCrunch AI · 2026-10-09
- sara questions an 'unsanctioned leak' accusation: name-and-shame or scapegoating? — suchenzang · 2026-10-09
- Peers vouch for fired OpenAI safety researcher Mikita: 'incredibly cracked' with high integrity — j_asminewang · 2026-10-09