Aegis
Run the following command in DeepSeek Harness:
dsh plugin install GanyuanRan/Aegis
Paste the following prompt into your AI chat to install this plugin:
In DeepSeek Harness, run dsh plugin install GanyuanRan/Aegis to install the Aegis method pack, with source at https://github.com/GanyuanRan/Aegis .
About this plugin
Imagine your AI coding agent making changes that drift from the project's intended contracts, leaving hidden pitfalls and forcing you to babysit every edit. Reworks pile up, and you never quite trust its "done". Aegis was built to change that — making your AI agent work like a disciplined engineer, not a carefree intern.
Aegis is a method pack that forces the agent to align with your real baseline — owners, contracts, and boundaries — before touching any code. After changes, it must prove completion with fresh verification evidence, covering scope and residual risks. It keeps simple tasks fast and adds ceremony only when complexity demands. It also tracks retired fallbacks and old paths, preventing silent technical debt.
On a frozen A/B benchmark, Aegis improved contract pass rate from 61.67% to 93.33%, and reduced unsafe outcomes from 13.33% to 0%. Best of all, it works across major hosts like Codex, Claude Code, OpenCode, and Kimi. If you're a developer or team that wants fewer reworks, safer changes, and trustworthy completion claims from your AI agent, Aegis is for you.
Screenshots
Use Cases
- Make AI coding agents understand your project baseline before editing, reducing blind changes
- Require agents to provide fresh verification evidence before claiming 'done'
- Keep simple tasks lightweight and apply heavier governance only for complex ones
Best For
- Developers frustrated by rework and untrustworthy AI agent deliveries
- Teams that need safer, verifiable AI code changes
- Managers seeking a consistent AI agent workflow across multiple hosts
Related Plugins
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.
Persistent, reusable multi-model Agent teams for DeepSeek Harness, with per-member model, role, tool policy, fallback routes, bounded DAG orchestration, quality gates, and a Run Center.