◆ NOISE IN → SIGNAL OUT◆ READALCHEMIST.COM◆ FREE / NO PAYWALL◆ NOISE IN → SIGNAL OUT◆ READALCHEMIST.COM◆ FREE / NO PAYWALL
THE DIGITAL ALCHEMIST
AIIMPACT 52

DeepSeek's Agent Repo Hit 21,000 Stars, Zero Merged Code

The dsh-superagent prototype rocketed up GitHub's trending charts in a week and stalled just as fast. Six days after launch, the repository still had exactly one commit.

2026-11-085 MIN READ#DeepSeek · #open source · #AI agents · #GitHub · #Claude Code
DeepSeek purple by RoadMaster19 (BY) via Openverse
DeepSeek purple by RoadMaster19 (BY) via Openverse

Twenty thousand nine hundred stars. One commit. If you want the whole story in two numbers, that's it.

On Saturday, Oct. 31, DeepSeek's Harness team dropped dsh-superagent-prototype on GitHub with no announcement beyond a WeChat post. It listens to GitHub and, optionally, Feishu. When someone asks the agent for work, it opens a work item, works on it in its own session and directory, and answers in the original conversation, for example with a pull request. MIT license, TypeScript, built on DeepSeek's existing Harness runtime. The agent uses the profile's default model unless overridden, working with any provider configured in Harness.

The stars came fast. By the next morning the repo had 4,137 stars and 381 forks. Bloomberg reported on Nov. 1 that DeepSeek is aiming it at Anthropic's Claude Code, as part of what the outlet describes as a bigger push by a lab that raised about $15 billion ahead of a planned IPO.

The curve already rolled over

Here's what the consensus misses: the growth already died. Trendshift's day-by-day tracker shows the repo gaining 2,024 stars on Oct. 31, then 4,952, then a peak of 5,884 on Monday, Nov. 2. After that it's a cliff: 3,613, then 1,881, then 928, then 761, and by Saturday, Nov. 7, just 43 new stars in 24 hours.

Daily GitHub Stars Gained, dsh-superagent-prototype
stars/day
02,0004,0006,000Oct 31Nov 2Nov 4Nov 6Nov 7
Source: Trendshift.io repository tracker, Oct. 31-Nov. 7, 2026

A repo that gains 43 stars a day isn't trending. It's done trending. Whatever verdict developers were going to render, they rendered it in four days.

The Digital Alchemist
The Digital Alchemist
dsh-superagent-prototype, Six Days In
20,900Total stars2,500Forks0Merged PRs intorepo1Commits sincelaunch
Source: GitHub repository metadata and MarkTechPost verification, Nov. 6, 2026

Stars don't merge code

Now the harder number: zero merged pull requests across every single day Trendshift tracked. MarkTechPost verified on Friday, Nov. 6 that the repository sat at one commit. Six days, 20,000 stars, 2,500 forks, and not one outside contribution landed in the codebase.

Some of that is by design. The prototype starts in observe mode for anything it was not explicitly asked to do and refuses to enable auto-merge from its own account, meaning a human must approve every change it proposes elsewhere. That's reasonable for a tool with full operating-system user access. But it also means the only performance numbers DeepSeek has published are simulated: a design document claims cutting simulated p95 task time to 15.7 minutes from 29.5, run against synthetic chat logs, not real repositories. As of early November, zero confirmed outside teams have run it against a live codebase.

The README is clear: "Research prototype for reference only, not an official DeepSeek product."

Translation: we built the plumbing, we're not promising it holds water.

The inference isn't dead. A model-agnostic, MIT-licensed agent that runs on a team's own infrastructure erodes vendor lock-in, if it works. DeepSeek has done this play before: the parent Harness repo counts 267,714 stars as of November 1, so the lab knows how to turn open code into mindshare. But mindshare isn't migration, and right now the evidence of actual migration is exactly zero.

What to watch: whether the repo gets a second commit, whether any team goes on record running it against a production codebase, and whether LangChain, CrewAI, or similar maintainers build an integration. If none of that shows up by Thanksgiving, treat the 21,000 stars as a trending-chart artifact, not a market signal.

Sources
  1. DeepSeek's deepseek-ai/dsh-superagent-prototype GitHub repository
  2. deepseek-ai/dsh-superagent-prototype — GitHub trending stats & insights
  3. DeepSeek Open-Sources Super-Agent, 0 Benchmarks
  4. DeepSeek Super-Agent Goes Open Source: $0, MIT
  5. DeepSeek releases the open source prototype of its Super-agent
  6. DeepSeek Open-Sources dsh-superagent, a Router-Worker Coding Agent
  7. DeepSeek dsh-superagent: MIT-licensed proactive coding agent
  8. [Research Prototype] Explaining DeepSeek's 'dsh-superagent'—A coding agent that monitors GitHub and submits PRs on its own|伊藤正章|サバイバルDXクリエイター:荒野のPip-Boy設計局
  9. DeepSeek's Super Agent Got 14k Stars. The Code Didn't Change - DEV Community
  10. DeepSeek Superagent Tutorial: Observe CI Before Repair
  11. DeepSeek dsh-superagent: Proactive GitHub Agent (MIT)
  12. DeepSeek's Super-agent: One Router, Many Workers
  13. DeepSeek Super-agent (dsh-superagent): Setup & Features
  14. DeepSeek Open-Sources GitHub Coding Agent That Files Its Own Pull Requests, With Security Warnings Attached
  15. DeepSeek Open-Sources Super-Agent, a Coding Agent That Works From Team Chat
  16. DeepSeek Releases Open-source Dsh-superagent Coding Agent - Dataconomy
  17. DeepSeek open-sources Super-Agent, a coding agent that reads team chats
  18. DeepSeek Takes Aim at Claude Code With Open-Source 'Super-Agent' That Can Work Without Constant Prompts
  19. DeepSeek Open Sources the Architecture Behind Its Super-Agent - Startup Fortune
  20. DeepSeek Super-agent: Open-Source GitHub Prototype Explained
  21. DeepSeek Super-agent: A Proactive Open-Source Coding Agent Prototype — What It Actually Is, and What It Is Not Yet - DEV Community
  22. DeepSeek Super-Agent: Open-Source Coding Assistant Launches
  23. Run DeepSeek's dsh-superagent on a VPS · SSD Nodes
  24. Dsh-superagent: The Coding Agent That Waits Its Turn
  25. DeepSeek Super Agent on GitHub: Setup, and Why Exit 3 Is Normal
← back to the feed
QueryFlow
ASK

Ask your database anything.

Stop writing queries. You don't need to know SQL to get the answer. QueryFlow AI writes it for you.

Native Mac app · by yForest

  • Snowflake
  • Redshift
  • PostgreSQL
  • MySQL
  • BigQuery
  • Databricks
  • Salesforce
  • Google Sheets
  • CSV/Excel
NVDA 150.96 ▼ 2.89%MSFT 508.29 ▲ 0.40%GOOGL 282.44 ▲ 0.80%AMZN 267.74 ▲ 3.82%META 616.38 ▼ 0.40%TSLA 383.72 ▲ 0.95%AMD 530.14 ▲ 5.31%AVGO 354.05 ▲ 3.64%PLTR 156.67 ▲ 0.07%COIN 110.51 ▼ 3.92%MSTR 90.32 ▼ 6.40%NVDA 150.96 ▼ 2.89%MSFT 508.29 ▲ 0.40%GOOGL 282.44 ▲ 0.80%AMZN 267.74 ▲ 3.82%META 616.38 ▼ 0.40%TSLA 383.72 ▲ 0.95%AMD 530.14 ▲ 5.31%AVGO 354.05 ▲ 3.64%PLTR 156.67 ▲ 0.07%COIN 110.51 ▼ 3.92%MSTR 90.32 ▼ 6.40%