Orchard Multi-Agent Research Kit

Important

Proprietary Orchard AI intellectual property. All Orchard AI-authored code, documentation, assets, prompts, configurations, designs, workflows, and other materials in this repository are proprietary to Orchard AI. They may not be used for independent work or for any purpose outside work expressly authorized by Orchard AI. Access to this repository does not grant a license. See LICENSE.

A workflow and small CLI used by Orchard AI to run multi-agent research with Claude or Codex, then consolidate the results into a local static report.

The kit is designed for work expressly authorized by Orchard AI. It can operate as a plain filesystem workflow or with Orchard session recovery, agent coordination, browser preview, and reflection patterns.

No domain-specific research data is included. The kit ships only generic templates, prompts, and workflow mechanics.

What It Does

• Scaffolds a research workspace with clear agent roles.
• Gives Claude/Codex agents copy-pasteable prompts for parallel research.
• Captures Orchard session context when the orchard CLI is available.
• Compiles agent markdown notes into a source-linked HTML report.
• Serves the report locally.
• Validates required sections, internal anchors, and workspace structure.

Quickstart

git clone <repo-url>
cd orchard-multi-agent-research-kit
npm test

node src/cli.js init ./runs/example --topic "Evaluate a product or technical decision"
node src/cli.js compile ./runs/example
node src/cli.js serve ./runs/example --port 19480

Open http://127.0.0.1:19480/.

Orchard-First Workflow

1. Create the workspace:

   orchard-research-kit init ./runs/my-topic --topic "Your research topic"

2. Use prompts/orchestrator.md in a parent Claude or Codex session.

3. Dispatch or open sibling agents using the generated files in agents/.

4. When Orchard is available, snapshot the live session state:

   orchard-research-kit orchard-snapshot ./runs/my-topic

5. Put each agent's finished notes in its assigned agents/*.md file.

6. Compile and validate:

   orchard-research-kit compile ./runs/my-topic
   orchard-research-kit validate ./runs/my-topic
   orchard-research-kit serve ./runs/my-topic

Why Orchard

Orchard makes this pattern stronger because multi-agent research is not just parallel prompting. It also needs:

• live session discovery
• transcript recovery after usage limits
• agent-to-agent coordination
• browser preview
• durable project context
• reflection and handoff records

This kit keeps those pieces visible without making Orchard mandatory.

Claude and Codex Usage

Use the same workspace with either tool:

• Claude: paste an agent brief into each session, then write results back to the assigned markdown file.
• Codex: run from the workspace root and let Codex read/write the assigned files directly.
• Orchard: use active sessions and messages for coordination, then run the compile/validate steps locally.

Commands

orchard-research-kit init <dir> --topic "..."
orchard-research-kit compile <dir>
orchard-research-kit validate <dir>
orchard-research-kit serve <dir> --port 19480
orchard-research-kit orchard-snapshot <dir>

After installation or npm link, orchard-research-kit is available as the CLI name. Before installation, use node src/cli.js.

Repository Safety

Before publishing a run, scrub it:

orchard-research-kit validate ./runs/my-topic
rg -n "secret|token|password|private" ./runs/my-topic

Do not commit proprietary research runs unless the topic owner has approved publication. The app and templates are proprietary Orchard AI materials, and run data may carry additional restrictions.

License

Proprietary. See LICENSE.