Aegis: Making DeepSeek Harness Coding Agents Work Like Disciplined Engineers
ganyuanran/aegis
Aegis makes DSH coding agents baseline-first, evidence-verified, and drift-checked for safer, rework-free changes.
Aegis provides a baseline-first, evidence-driven coding methodology for DeepSeek Harness, reducing rework and improving change safety.
dsh plugin --profile web add github:ganyuanran/aegis- Category
- Development & Operations
- Platform
- DSH-Plugin
- Author
- ganyuanran
- Distribution
- Plugin
aegis Key Features
aegis Repository
aegis Repo Summary
Aegis is a method pack plugin (DSH plugin) for DeepSeek Harness that makes AI coding agents work like disciplined engineers, reducing rework, improving change safety, and providing verifiable evidence before tasks are marked done. It addresses the core problem that AI agents often lack understanding of a project's real baseline before editing code, leading to guesswork, unsafe changes, and blind 'done' claims. Aegis uses baseline-first, evidence-driven, and drift-checked methods to ensure agents plan against the actual project baseline before editing, prove completion with fresh evidence, and keep simple tasks simple. Maintained by GanyuanRan under the MIT license, written in Python, and last updated in August 2026, the plugin has over 1000 stars on GitHub and supports popular coding agents like Claude Code, Codex, and OpenCode.
What core features does the Aegis DSH plugin offer?
Aegis provides a comprehensive method pack with key features: baseline alignment (agents learn project owners, contracts, and boundaries before touching code), A/B benchmark testing (measured on a frozen held-out set, contract pass rate improved from 61.67% to 93.33%, unsafe outcomes from 13.33% to 0%), evidence verification (completion claims include fresh verification evidence, covered scope, and residual risk), ghost code cleanup (retired fallbacks and old paths are tracked or removed via retirement triggers to prevent technical debt), and a fast path (trivial requests are handled directly, with ceremony only appearing for complex tasks). These features make agents more trustworthy and reduce human supervision.
How do I install the Aegis DSH plugin?
Installing Aegis is straightforward. Run the following command in your DeepSeek Harness environment to add the plugin:
dsh plugin --profile web add github:ganyuanran/aegis
After installation, the plugin automatically registers its skills and tools, and you can enable it in your agent configuration. Refer to the Fast-Track Playbook in the repository for quick onboarding.
Which DeepSeek Harness versions and platforms are compatible with the Aegis DSH plugin?
As a DSH plugin, Aegis integrates seamlessly with the DeepSeek Harness framework and supports multiple mainstream AI coding agents, including Claude Code, Codex, and OpenCode. It interacts with the filesystem via the dsh-skill-filesystem dependency, ensuring cross-platform consistency. Being written in Python, it runs on any environment supporting Python (e.g., Linux, macOS, Windows). The project is actively updated, so keeping the plugin version current is recommended.
What is the license and community support for the Aegis DSH plugin?
Aegis is licensed under MIT, allowing free use, modification, and distribution. The project is maintained by GanyuanRan, with 1022 stars and 46 forks on GitHub, indicating an active community. Discussions and articles are available on LINUX DO forums and DEV.to, and project details can be found on olud.ai. Last updated in August 2026, the project remains actively maintained.