AI Agent Hub
Back to plugins
⚙️

dsh-subagent-error-details

Workflow Updated 2026.08.26

Run the following command in DeepSeek Harness:

dsh plugin install sijie-ni-0214/dsh-subagent-error-details

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

In DeepSeek Harness, run the following command to install (source: https://github.com/sijie-ni-0214/dsh-subagent-error-details): dsh plugin install sijie-ni-0214/dsh-subagent-error-details

About this plugin

When a background subagent dies mid-turn, the core DSH settlement notice prints a single fixed sentence and throws away the real failure reason entirely. Whether the cause is a 429 rate-limit, a timeout, or a logic error, the parent agent only sees "subagent finished" with no diagnostic detail and no way to make a meaningful retry or fallback decision. dsh-subagent-error-details exists to close that information gap.

The plugin subscribes to the public subagent/end event as a host-plane bundle. When stopReason is error, it resolves the child session terminal failure detail through sessionPersistence.load(), reading from the live session store first and the durable JSONL log second, then routes a short companion message back to the owning parent via the session-header parentSession chain and the agent registry. Every code path is wrapped defensively: a plugin bug degrades to "no details" rather than breaking the parent loop, and all service lookups are deferred inside the handlers, so activation can never fail a session mount. There are zero runtime dependencies.

This plugin is aimed at developers running DSH multi-agent workflows where background subagents do the heavy lifting and the parent needs to know why a child failed. It works across every agent preset with no composition editing required. Once the upstream core fix ships the error text directly in the official notice, the plugin auto-detects it and suppresses the companion message, so no manual removal is needed.

Use Cases

  • A background subagent dies from a 429 rate-limit and the parent needs the exact error code to decide between retry and fallback
  • In multi-agent orchestration a long-running child errors out and the parent must distinguish timeout from rate-limit from a logic bug to pick the right recovery path
  • While debugging a complex agent chain, pinpoint the child terminal-turn failure root cause without re-running the entire scenario

Best For

  • Developers building DSH multi-agent workflows that rely on background subagents for delegated tasks
  • Engineers who need the parent agent to make retry or fallback decisions based on child failure reasons
  • Technical staff debugging agent chains and investigating unexpected subagent behavior