DeepSeek releases open-source Super-Agent, an always-on AI coding agent rivaling Claude Code

·12 views

DeepSeek's Harness team has released Super-Agent, an open-source AI coding agent architecture on GitHub under the MIT license, positioned as a competitor to Anthropic's Claude Code. Unlike Claude Code's terminal-driven workflow, Super-Agent runs as a background service that monitors chats, GitHub pull requests, and CI/CD pipelines, and it can act on events without being explicitly invoked. Its design splits work between a lightweight Router that triages triggers, Workers that execute coding tasks in parallel, and a Watchdog that prevents stalled or looping tasks. DeepSeek's synthetic benchmarks reported a 95th-percentile completion time of 15.7 minutes, versus 29.5 minutes for traditional single-session subagent approaches. The system also includes permission tiers, including an observe mode that logs intended actions without executing them.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Designing for Future Trends, Not Just Today's Problems

A writer reflects on a Forbes article describing twelve mega themes shaping the future, including connected living, intelligent assistants, digital reality, autonomous systems, and zero-latency technology. The author argues that breakthrough solutions often emerge from examining how these trends may intersect rather than focusing only on current issues. They suggest designing for how users may work five or ten years ahead, instead of how they work today. Applied to game development, this approach could mean moving beyond basic merge-conflict tools toward collaboration systems that use AI, real-time feedback, and connected environments. The core takeaway is that effective solutions should solve present problems while being prepared for the direction the world is heading.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Navy Veteran Begins Public Log of Path to Becoming a Software Engineer

Brandon, a U.S. Navy veteran and computer science student, has started documenting his journey toward a software engineering career on DEV Community. He is studying at Mt. San Jacinto College with plans to transfer to a four-year university, and is currently focused on C++ while also exploring Python, Unreal Engine, and Arduino. His long-term goals include working in game development or aerospace, with a particular aim of one day joining NASA. The series will cover his coding struggles, projects, internship and interview preparation, and lessons learned, with the stated aim of showing beginners the unpolished reality of learning to program.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

FoodBridge Triangle Wins First Prize at Hack for Humanity Raleigh

Developers Desmond Gatling and the author built FoodBridge Triangle at Hack for Humanity: Raleigh to simplify finding local food assistance. The tool lets users search a directory of grocery services, prepared meals, and benefits resources, view locations on a map, and read published eligibility requirements. Users can bookmark resources, build a printable plan, and ask a Gemma-powered Community Helper questions in plain language. The prototype starts with seven curated Raleigh resources and explicitly does not approve or deny assistance, instead directing users to confirm details with providers and official sources.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Resetting Mock Servers Between Tests Prevents Order-Dependent Failures

Stateful mock servers that persist stubs, counters, and scenario positions across tests can make suites fail for reasons unrelated to the code under test. Real isolation means returning the mock to a known baseline before each test, either by using a disposable instance, calling an explicit reset, or scoping each test to a unique context. Resetting in the setup phase rather than in cleanup ensures a failed test cannot corrupt the tests that follow. A test suite that only passes when run in a specific sequence has hidden dependencies that should be removed. Documenting the baseline configuration in a versioned file helps new contributors understand the mock's starting state.

0 comments Read more at DEV Community