Project
MIT-licensed, plugin-based open-source runtime for building coding agents, released at v0.1 as a developer preview with a Python SDK and local web interface.
DeepSeek Harness, DeepSeek's open-source agent runner, carries a 9.4-severity flaw that lets a sandboxed agent grant itself full access with one command. Its control API has no authentication and trusts a client-supplied Host header.
Reality
DeepSeek released its open-source Harness agent runtime as a Windows and macOS desktop app on September 30, so users no longer need a command line to run it. The agent edits local files and runs commands, so what each user exposes to it decides what it can touch and which provider sees it.
Perspective Coverage
3 publishersReality
Harness v0.1 shipped under MIT on the same day V4-Pro went generally available, three days before peak pricing lands. The lock-in it targets is the runtime, not the weights.
Perspective Coverage
4 publishersReality
The MIT-licensed dsh runtime bundles sessions, tool calls, permissioning and a local web UI, with model adapters as plugins covering Anthropic, OpenAI, Bedrock, Vertex and Azure. It is still a preview.
Reality
A post from NVIDIA's AI safety and security teams cites three summer reports of frontier agents leaving their boundaries, and argues only infrastructure can hold final authority.
Reality
AWS has open-sourced a preconfigured general-purpose agent on top of its Strands SDK and says it runs 45% cheaper than Claude Code and Codex. The model call defaults to Amazon Bedrock, and Marc Brooker says one line changes that.
Reality
The Unified Harness Protocol specifies how an application starts a task on an agent runtime, follows it, cancels it and collects the files, borrowing the shape of OpenAI's Responses API so existing streaming clients need no changes.
Reality
A Rust daemon and a Chromium extension open a separate window inside the developer's daily browser profile, so the agent inherits every live login. The documented approval prompt covers tabs you already have open.
Reality
The DeepDeck benchmark runs one task twice, once with the WebMCP bundle installed and once with the tool list forced empty, freezing the model configuration in between and leaving correctness unscored unless you supply an answer you checked yourself.
Reality
The installer launched fine on the build machine and quit in about a second on a Mac that had never run Homebrew. Getting it to start took 26 dylibs copied into the bundle, every path rewritten to @loader_path, and an ad hoc re-sign.
Reality
The V4.1-Flash change log puts Terminal Bench 2.1 at 90.6 against V4-Pro's 87.9 and DeepSWE at 74.2 against 62.7. DeepSeek says API prices came down with the release and points to a pricing page for the amounts.
Reality
OX Security found that DeepSeek's coding harness would let an agent unlock itself simply by asking, because the gate trusted a name the caller supplies. That makes agent isolation a claim you have to attack before you report it as a control.
Reality
VulnCheck rates CVE-2026-82533 at 9.4. DeepSeek pushed a fix on August 27. The build named in the record never reached npm, and the sandbox itself still confines only writes.
Reality
A dev.to writeup attributes 612.9M tokens and $28.35 to four days of DeepSeek Harness work on a C# arbitrary-precision library, and the part worth reading is how the agent checked its own rewrite.
Reality
The release attaches speculative decoding to weights it publishes under MIT, which lowers what a self-hosting threat costs to stand up, even though the performance claim behind it is still the vendor's own.
Reality
DeepSeek Harness is an MIT-licensed developer preview in which the model adapter, tool registry, session log and agent loop are all plugins. The interesting part is what that implies for evaluation.
Reality
Four subsystems are all Cordis plugins at the pinned commit, which turns extension into an architecture decision. The same guide records GitHub at rc.8 and npm at rc.7.
Reality
An independent recount of six usage trackers found gaps of 2.00x to 8.09x. The bigger problem is that none of them, and no invoice, can say which number was right.
Reality
A viral X post said an inference-time text layer put DeepSeek V4 Pro ahead of Fable 5 on every task. The report it points to shows single runs, nine benchmarks, and two losses.
Reality