AI Agent Hub
Back to plugins
⚙️

dsh-shadow-mind

Workflow Updated 2026.09.04

Run the following command in DeepSeek Harness:

dsh plugin install whutzefengxie-ops/dsh-shadow-mind

Paste the following prompt into your AI chat to install this plugin:

Run the install command in a DeepSeek Harness terminal; the full source repository is available at https://github.com/whutzefengxie-ops/dsh-shadow-mind , and once installed you can configure Shadow agent definitions and inspect lifecycle logs in the Web UI.

About this plugin

In a multi-agent DeepSeek Harness workflow, the root agent often finishes a tool turn without any independent asynchronous review confirming the result was sound. dsh-shadow-mind fills that gap: after each eligible tool-using turn it spawns background Shadow agents in isolated child sessions, lets them inspect the context and emit structured findings with verdict, severity, and reference anchors, then relays accepted reports back into the root session—delivering a main-thread-focuses, side-thread-quality-checks collaboration model.

The plugin writes every lifecycle event (admission, child start, cancellation request, terminal outcome, report delivery) into JSONL logs tagged with stable reason codes such as SHADOW_TIMEOUT or PROVIDER_ABORTED. The default trajectory projection strips reasoning text, tool arguments, credentials, and absolute paths, retaining only length and a SHA-256 digest to balance observability with privacy. A zero-dependency companion tool, shadow-debug.mjs, rebuilds the full run context from a single child session ID or error fragment pulled from local logs.

Built for developers assembling multi-agent pipelines on DeepSeek Harness who want an async quality gate around critical tool calls, backed by a lightweight, auditable debugging trail.

Use Cases

  • Spawns isolated Shadow agents after each eligible tool turn to review results and relay structured verdicts back to the root session
  • Pinpoints Shadow timeouts, cancellations, or delivery failures via JSONL lifecycle logs tagged with stable reason codes
  • Runs parallel quality checks in child sessions without blocking the root agent, ideal for long multi-tool workflows

Best For

  • Developers building multi-agent pipelines on DeepSeek Harness who need an async quality gate
  • Teams adding auditable review steps around critical tool calls without blocking the root agent
  • Ops engineers requiring a lightweight debugging trail and privacy-preserving log projection