Zurück zu den Neuigkeiten
ProduktAI Understanding Briefing

DeepSeek Harness fügt experimentelle Claude Code Mods-Kompatibilitätsebene hinzu

DeepSeek veröffentlichte Harness v0.2.1-alpha.1 und führte eine experimentelle Kompatibilitätsschicht ein, die es Claude Code-Mods ermöglicht, innerhalb seines Plugin-Systems zu laufen, und demonstrierte die Wiederverwendung von Erweiterungen über mehrere Tools hinweg.

4 min readRead the linked source
Source-provided image accompanying DeepSeek Harness adds experimental Claude Code Mods compatibility layer
QuellenangabeQuelle aufgezeichnet
Herausgeber
eu.36kr.com
Quelllink
eu.36kr.comhttps://eu.36kr.com/en/p/4010809598922624
Quelltyp
Verknüpfte Quelle – Der Status der Primärquelle wurde nicht festgelegt.
KontextVerstehen Sie dies in 60 Sekunden

Beginnen Sie hier

Schlüsselbegriffe

API (Anwendungsprogrammierschnittstelle)
Eine strukturierte Möglichkeit für ein Softwaresystem, Anfragen an ein anderes System zu senden und Antworten von diesem zu empfangen.
KI-Agent
Ein Softwaresystem, das beobachten, schlussfolgern und Maßnahmen ergreifen kann, um ein Ziel zu erreichen, häufig unter Einsatz von Werkzeugen und Gedächtnis.
Funktion
Eine Eingabevariable, die von einem Modell verwendet wird, um Vorhersagen zu treffen.
Testen Sie sich selbstKI-Agenten-Quiz

Was ist passiert?

DeepSeek released version 0.2.1-alpha.1 of its DeepSeek Harness (DSH), an open-source agent framework. This update introduces an experimental compatibility layer for Claude Code Mods, allowing extensions written for Anthropic's Claude Code to run within DSH's plugin architecture. The release also includes a new 'Let Agent create plugins' and a developer toolkit for debugging. According to 36Kr, the update was signed by Cui Tianyi, head of the DeepSeek Harness team, who previously commented on the convergence of plugin concepts between the two platforms.

DeepSeek released DeepSeek Harness (DSH) version 0.2.1-alpha.1, which includes an experimental compatibility layer for Claude Code Mods. This layer allows extensions developed for Anthropic's Claude Code interface to run within DSH's plugin system. The release notes, signed by Cui Tianyi, state that the primary goal is to verify that Claude Code Mods API capabilities are a subset of DSH's plugin capabilities.

The update also introduces a user-friendly entry point called 'Let Agent create plugins,' allowing users to describe desired features in natural language for the agent to generate and install plugins on the fly. A supporting developer toolkit was added to help users view session logs, locate issues, and debug plugin behavior.

According to 36Kr, Cui Tianyi had previously commented on the release of Claude Code Mods, noting that the concepts of the two systems were converging. He emphasized that while Claude Code opens specific capabilities as Mods, DSH was designed with an 'Everything is a Plugin' philosophy, where models, tools, sessions, and UI elements can all be treated as interchangeable plugins.

The compatibility layer currently supports bridging examples such as 'Token Weather' (context usage visualization), 'Blast Radius' (pre-execution impact analysis), and 'Replay Theater' (step-by-step file modification review). However, the documentation explicitly states that full practical compatibility is not yet available, with features like built-in diffs, agents-md, sec-default, and telemetry marked as non-runnable due to unconnected dependent events or interfaces.

Quellenangaben: eu.36kr.com ↗

Warum es wichtig ist

This update signals a shift toward interoperability in the tooling ecosystem. By demonstrating that Claude Code Mods can function as a subset of DSH's broader 'Everything is a Plugin' architecture, DeepSeek highlights the potential for a shared extension standard. This reduces vendor lock-in for developers who build custom AI workflows, allowing them to reuse components across different agent harnesses. It also validates the modular design of DSH, showing it can accommodate external interfaces without core architectural changes.

The move demonstrates a practical step toward interoperability in the ecosystem, which has historically been fragmented by proprietary tooling. By showing that extensions from one major platform (Claude Code) can run on another (DeepSeek Harness), the update suggests that a common extension standard is feasible.

This reduces the cost of switching between frameworks for developers and enterprises. If plugins can be reused, organizations are less locked into a single vendor's ecosystem, fostering competition based on core model and harness performance rather than proprietary extension markets.

The 'Everything is a Plugin' architecture of DSH is validated by this update, showing its flexibility to accommodate external interfaces. This modular approach may become a competitive advantage for DeepSeek in the open-source agent tooling space, attracting developers who value customization and portability.

The experimental nature of the release indicates that the industry is still in the early stages of defining standards for agent customization. The success of this compatibility layer could influence future API designs from other AI providers, potentially leading to a more unified approach to agent extension development.

Interactive Mechanism

Interaktiver Mechanismus: Wie es tatsächlich funktioniert

Entdecken Sie interaktiv die zugrunde liegende Technologie, die dieser Entwicklung zugrunde liegt.

Agent Lifecycle Stage:
1
User Intent & Planning: "Audit customer refund request #4092 and settle payment."
2
Tool Calling: Emits structured JSON call crm_get_transaction(id='4092').
3
Guardrail & Verification:🛡️ Paused: High-value action requires human operator sign-off.
4
Final Settlement: Refund recorded, email receipt dispatched, and audit log stored.
Core takeaway: An AI agent is not just a language model—it is a closed loop of planning, tool invocation, and environment feedback. Production systems require self-healing retries and strict human approval guardrails.
Interaktiver Konzeptcheck+10 Points
AI Agents Quiz

An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?

Was Sie als nächstes sehen sollten

Monitor whether this experimental layer evolves into full practical compatibility, including support for currently non-runnable features like built-in diffs and telemetry. Watch for community adoption of cross-tool plugins and whether Anthropic or other major AI tool providers respond with similar interoperability standards. Additionally, track the development of the 'Plugin Engineering' concept as it may become a standardized practice in customization.

Watch for updates to the DSH compatibility layer that address the currently non-runnable features, such as built-in diffs and telemetry. Full compatibility would significantly increase the utility of the cross-tool plugin ecosystem.

Monitor community response and adoption of the 'Let Agent create plugins' . If users successfully generate and share plugins across different harnesses, it could accelerate the development of a shared plugin library.

Observe whether Anthropic or other AI tool providers respond to this interoperability move. A collaborative approach to standardizing agent extensions would benefit the entire industry, while a competitive response could lead to further fragmentation.

Track the evolution of the 'Plugin Engineering' or 'Mod Engineering' terminology and practices. As these concepts mature, they may become standard job roles or development practices in the AI industry.

Verwandte Leitfäden und Quizze

KI-AgentenKI-Modelle erklärtPrompt EngineeringTesten Sie, was Sie wissen – probieren Sie ein kostenloses KI-Quiz ausSuchen Sie in unserem Glossar nach einem KI-BegriffFolgen Sie dem AI-Modell-Release-Tracker
Fanden Sie das nützlich?