SponsorLobeHubLobeHubLearn more
dshfind
Back to plugins

A method pack that makes AI coding agents trustworthy: they align with your project's real baseline — owners, contracts, boundaries — before editing, and ship completion claims with fresh verification evidence. On a frozen A/B benchmark, contract pass rate went 61.67% → 93.33% and unsafe outcomes 13.33% → 0%.

Star
1,013
7-day growth
Contributors
2
Language
Python
Updated
2026-08-15
Install
Derived from package metadata
dsh plugin --profile web add aegis

The name after --profile is a profile name you choose yourself — swap in whichever one you use.

Editor's notes

Aegis is aimed squarely at the "I can't stop babysitting my agent" problem. It isn't a plugin or a daemon — it's a method pack. Before touching code, the agent aligns with your project's real baseline: who owns what, what the contracts are, where the boundaries sit, instead of guessing. When it claims something is done, that claim ships with fresh verification evidence, the scope it covered, and the residual risk — you read evidence, not vibes. Retired fallbacks and old code paths get tracked or removed under a retirement trigger, so technical debt stops accumulating silently. It also deliberately keeps a fast path: trivial requests stay trivial, and ceremony only shows up when the task genuinely calls for it. The effect was measured on a frozen held-out A/B benchmark (Aegis 2.7.6, 120 valid runs across 20 cases, both arms on gpt-5.6-sol / xhigh): contract pass rate rose from 61.67% to 93.33% and unsafe outcomes fell from 13.33% to 0%. The same pack runs on Codex, Claude Code, OpenCode, Kimi and DeepSeek Harness, with a dedicated DSH setup guide.

  • Aligns with the project's real baseline — owners, contracts, boundaries — before any edit, so there is less rework and less blind trust
  • Completion claims arrive with fresh verification evidence, covered scope and residual risk — evidence, not vibes
  • Frozen held-out A/B benchmark: contract pass 61.67% → 93.33%, unsafe outcomes 13.33% → 0% (bounded advisory evidence)
  • Simple tasks stay simple; ceremony appears only when needed. One pack spans DSH, Codex, Claude Code, OpenCode and Kimi
Score breakdown
Scored 2026-08-15
Activity
20.1
Star heat
28.6
Engineering
20.9
Maintainer
5.7

让 AI 编码 agent 架构感知的方法包(baseline/证据/防漂移),有 dsh 字段并依赖 @deepseek-ai/dsh-skill-filesystem,但本质是跨 agent 通用 method-pack,DSH 只是承载目标之一。

Share card

Drop it into your README — it links back to this page.

Badge (inline)
Badge (inline)
[![dshfind](https://dshfind.com/api/badge/GanyuanRan/Aegis)](https://dshfind.com/en/plugins/GanyuanRan/Aegis?ref=badge)
Card (440×200)
Card (440×200)
[![dshfind](https://dshfind.com/api/card/GanyuanRan/Aegis)](https://dshfind.com/en/plugins/GanyuanRan/Aegis?ref=badge)