dsh-gentle-engram
用于 DeepSeek Harness 中 Engram 持久化内存的 Cordis 主机集成插件
AI 分析
核心用途是为 DSH 接入 Engram 持久化内存生命周期管理。适合需要记忆上下文、跨会话保持记忆的 AI 代理任务。必要条件是系统已安装 Engram 并可用。
安装
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:eehcx/dsh-gentle-engram说明文档
阅读完整 README ↗dsh-gentle-engram
Host Cordis plugin written in TypeScript that adds reliable Engram lifecycle behavior to DeepSeek Harness.
This integration requires Engram to be installed and available on your system; the bridge cannot work without it.
Design
- Uses the existing
@deepseek-ai/dsh-mcp-clientrow for official Engram MCP stdio tools. - Hooks DSH session start, inbound prompts, tool results, turn stopping, and agent disposal.
- Seeds relevant Engram context into the next model step.
- Saves prompts and filtered passive tool learnings asynchronously.
- Serializes per-session writes and waits for pending writes at turn boundaries.
- Closes Engram sessions and records a structured summary when an agent is disposed.
- Never accesses Engram SQLite directly and avoids forwarding likely secrets.
Installation from npm
Requirements:
- DeepSeek Harness with the
dsh plugincommand. - Engram installed and available as
engraminPATH. - A configured DSH profile, such as
web.
Install the alpha channel with:
npx @deepseek-ai/dsh plugin --profile web add dsh-gentle-engram@alpha
For the stable channel:
npx @deepseek-ai/dsh plugin --profile web add dsh-gentle-engram
Restart DSH after installation. The package is a Host bundle: its cordis.patch.yml owns both the MCP bridge and lifecycle plugin. Do not edit shipped presets or add a second manual Engram MCP row.
Check that Engram is available before starting DSH:
engram --version
Local development
pnpm install
pnpm run typecheck
pnpm run build