Add Parallel Responses research subagent plugin for DeepSeek Harness

[georgeatparallel]

What this adds

This adds a parallel_research tool to DeepSeek Harness.

Think of web_search as asking for useful pages to look through. parallel_research is more like asking a research assistant to look into a question and come back with an answer and the sources behind it.

The two tools work side by side. The agent can still use web_search when it wants to inspect search results itself, or use parallel_research when it wants to hand off the research.

How it works

1. The agent sends a complete research question to parallel_research.
2. The plugin passes that question to the Parallel Responses API.
3. Parallel researches the question and returns an answer with citations.
4. The agent gets the answer back and decides what to do next.

The agent can also ask several independent questions at the same time. DeepSeek Harness already manages those tool calls, so the plugin does not add its own scheduler or concurrency limit.

Research depth can be set to low, medium, or high. medium is the default. low is useful when a problem can be split into smaller questions and researched in parallel.

Only the delegated question and the plugin's fixed research instructions are sent to Parallel. Conversation history, local files, workspace state, and other tools are not included automatically.

Why this is useful

An agent should not have to run every search, read every page, and piece together every answer itself. It can hand a question to a research specialist, get back a useful answer with sources, and stay focused on the larger task.

That becomes especially helpful when several independent questions can be researched at once.

The package is not published to npm yet, so the README explains how to install it from a local tarball.