MIT 授權的 AI 程式設計代理可靠性方法包，可透過 `dsh plugin add github:GanyuanRan/Aegis` 作為 profile 外掛安裝進 DeepSeek Harness（dsh）。
DSH 整合
相容
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
MIT
01它能幫你完成什麼？
Make AI coding agent runs in DeepSeek Harness safer and less prone to rework
The agent plans against the project's real baseline before editing and completion claims ship with fresh verification evidence, covered scope and residual risk — on the project's frozen A/B benchmark, contract pass rate went 61.67% → 93.33% and unsafe outcomes 13.33% → 0%
Developers running coding agents (DeepSeek Harness, Codex, Claude Code, OpenCode, Kimi) who want fewer reworks and safer changes on non-trivial tasks
Apply one discipline across every AI coding host you use, including DeepSeek Harness
The same method pack installed on Codex, Claude Code, OpenCode, Kimi, DeepSeek Harness and other skill-aware hosts, with host-aware updates via `scripts/aegis-update.py`
Teams standardizing agent workflows across multiple coding-agent hosts
02如何將外掛接入 DeepSeek Harness？
先決條件
- DeepSeek Harness (`dsh`) installed and a profile selected — installation is a profile-plugin install via `dsh plugin --profile <profile> add github:GanyuanRan/Aegis`
- Python available to run the verification scripts (`scripts/aegis-doctor.py`) from the installed method-pack root
安裝步驟
- 01
Give the README's quick-install prompt to your agent: read the Aegis repo, identify the host (DeepSeek Harness / dsh), and install via `dsh plugin --profile <profile> add github:GanyuanRan/Aegis` instead of silently substituting the direct-child compatibility path
$ dsh plugin --profile <profile> add github:GanyuanRan/Aegis
- 02
Restart or reload the host if needed
- 03
Locate the installed `<aegis-method-pack-root>`, then run `cd <aegis-method-pack-root> && python scripts/aegis-doctor.py --write-config --json` — not from the target project directory
- 04
If the host uses a separate skill discovery directory, verify it with `--discovery-root <path>`; if the host guide declares a skill directory name prefix, also pass `--discovery-name-prefix <prefix>`
- 05
Complete the DeepSeek Harness host guide's native activation and automatic-entry checks (file discovery or a generic doctor result alone is not sufficient when the host provides a plugin, hook, or session-start bootstrap contract)
驗證整合成功
- Treat the install as complete only if the doctor JSON includes `"ok": true`, `"workspaceSupport": "available"`, and `"configStatus": "configured"`
03DSH 整合程度與能力邊界
Installed into DeepSeek Harness (dsh) as a profile plugin via `dsh plugin --profile <profile> add github:GanyuanRan/Aegis`, with a dedicated DeepSeek Harness host guide
Baseline-first planning
coding tasks in any supported host, including DeepSeek Harness→the agent aligns with the project's real baseline — owners, contracts, boundaries — before touching code
lazily creates CONTEXT.md / CONTEXT-MAP.md files on the first resolved domain term (glossary-only)Evidence-based completion verification
completion claims from the agent→fresh verification evidence, covered scope and residual risk attached to every "done" — you read evidence, not vibes
Install & activation management (doctor / update)
the installed method-pack root→`scripts/aegis-doctor.py` reports install status and switches activation mode (`auto`/`explicit`) and TDD mode (`off`/`auto`); `scripts/aegis-update.py` performs host-scoped updates
doctor runs with `--write-config` during install verification; the host must be restarted after changing activation modeExplicit workflow triggers
natural language or explicit requests in the host session→`Aegis goal: ...` frames scope, success evidence and boundaries; `Grill me ...` starts a decision interview; `TDD Route: strict` / `test-first` requests strict test-first work; `aegis:first-principles-review` pressure-tests a direction
04適合誰？何時不該用？
適合
- Developers running coding agents (DeepSeek Harness, Codex, Claude Code, OpenCode, Kimi) who want fewer reworks and safer changes on non-trivial tasks
- Teams standardizing agent workflows across multiple coding-agent hosts
不適合
- For DeepSeek Harness an install guide exists, but release-level fresh host smoke is still pending — the integration is documented, not yet verified per release.
05相容性、維護與安全提醒
- Aegis is the runtime-ready Method Pack, not the full Aegis Platform, a daemon, a background runner, a runtime core, or final completion authority; user instructions and target-project rules outrank Aegis guidance.
- For DeepSeek Harness an install guide exists, but release-level fresh host smoke is still pending — the integration is documented, not yet verified per release.
- The benchmark numbers are bounded advisory evidence from a frozen held-out A/B benchmark, not a universal-quality or completion-authority claim.
MIT · actively maintained (latest release v2.8.4, 2026-08-19)
06常見問題
Aegis 如何接入 DeepSeek Harness？原生還是 MCP？
不是 MCP 方式：對官方 DeepSeek Harness（`dsh`），全域／最小安裝即原生 profile 外掛安裝，指令為 `dsh plugin --profile <profile> add github:GanyuanRan/Aegis`，依專門的 DeepSeek Harness 主機指南操作。只有當外掛管理工具不可用且你明確核准相容模式時，才退回 direct-child 相容路徑。
如何驗證安裝成功？
在安裝後的方法包根目錄（不是你的專案目錄）執行 `python scripts/aegis-doctor.py --write-config --json`。只有當 JSON 中包含 `"ok": true`、`"workspaceSupport": "available"`、`"configStatus": "configured"`，並完成主機指南的原生啟用與自動進入檢查，才算安裝完成。
需要什麼前置條件？
已安裝 DeepSeek Harness（`dsh`）並選定一個 profile 用於安裝，以及用於執行 doctor/update 腳本的 Python。Aegis 本身不要求額外的 API key 或帳號。
有什麼限制？Aegis 對專案有最終決定權嗎？
Aegis 目前是執行時期就緒的 Method Pack，不是完整平台、常駐程式或最終完成權威；使用者指令與目標專案規則優先於 Aegis 指引，基準數字僅為有邊界的參考證據。對 DeepSeek Harness 主機，發布級新鮮冒煙驗證仍在等待中。
之後如何更新或切換模式？
輸入 `update Aegis` 或 `aegis:update` 即可走主機感知的更新路徑（`scripts/aegis-update.py`）。啟用模式預設為自動，可用 `python scripts/aegis-doctor.py activation-mode explicit` 切換為明確模式並重啟主機；TDD 預設關閉。
07相關的 DSH 工作流程
distilly
作者 titanwings
Distilly — Distill how they think into reusable Skills for any Agent or Bot. Formerly Colleague Skill（原同事 Skill）.
dsh-market
作者 dsh-market
The plugin market inside DeepSeek Harness — browse, search, one-click install · DSH 視覺化外掛市場
ai_animation
作者 unclecheng-li
本專案整理了用於生成[炫酷 HTML 動畫網頁]的 AI Prompts，涵蓋動畫效果、3D 視覺化、PPT 風格演示、UI 美化等多個類別。
skillcorpus
作者 evermind-ai
開源智慧體技能庫基礎設施：把零散的 SKILL.md 檔案整理成可檢索、即取即用的技能語料庫，並附帶檢索與評估工具。
08資料與來源
If the host is the official DeepSeek Harness (`dsh`), treat global/minimal installation as native profile-plugin install…
`DeepSeek Harness` | Install guides exist; release-level fresh host smoke is still pending | [DeepSeek Harness](docs/REA…
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-21。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。