Check out our open source tools to give Claude Code GTM superpowers→

Research Process Builder

Factory that produces portable research agent prompts at 90-100% accuracy.

Open source · MIT · Production-tested inside LeadGrow's live pipeline

Quick install

$ git clone https://github.com/MitchellkellerLG/research-process-builder.git

View on GitHub →Want a custom version?

What you unlock

A self-annealing methodology that turns any web-research goal into a validated, step-by-step process. Generates 15-20 candidate search patterns, tests them against companies across three size tiers, scores quality and consistency, and iterates until 90%+ reliability. The output is a portable markdown file any agent — Claude Code, Claygent, custom GPT, OpenAI Agents — can follow to surface specific intelligence reliably.

OutcomeBuilt at LeadGrow, used in live pipelines, shipped open source so you don't have to build it yourself.

How it works

See it in action

3 things you can do right now.

No UI. No clicking. Just commands that execute.

Generate candidate patterns for a research goal

$ research-builder generate

--goal "find recent funding rounds"

--company notion.com

✓ 18 candidate patterns generated Testing against notion.com...

Outcome18 search strategies in 30 seconds. No manual brainstorming. The loop does the work.

Test patterns across company tiers

$ research-builder test

--task "funding-rounds"

--companies companies.csv

--tiers enterprise,mid,smb

✓ 9 companies tested across 3 tiers → 16/18 patterns pass ≥90%

OutcomeYou know which patterns hold before deploying them at scale. No guessing.

Inspect the validated output

$ research-builder inspect

--task "funding-rounds"

--format markdown

✓ Process: funding-rounds.md → 4 validated steps → 3 kill-list exclusions

OutcomeA portable process any agent can follow. Drop it in Clay, Claude Code, or a custom GPT.

What's included

Every capability, ready to script.

9 production processes: competitors, reviews, news, hiring, growth signals, and more

Validated across 220+ patterns and 11 companies from SpaceX to micro startups

Kill lists cut wasted searches by 30-40%

Portable `.md` output works in Claude Code, Clay, custom GPT, OpenAI Agents

6-phase methodology for building new processes for any research goal

The workflow

End to end. Zero manual steps.

9 production processes validated across 220+ patterns. Drop the .md into any agent.

01

Define the research goal.

Name the goal and describe what a good answer looks like. The builder needs a success criterion to score against — be specific about what you're trying to surface.

$ research-builder init

--goal "find LinkedIn headcount changes"

--name "headcount-tracking"

02

Generate candidate patterns.

The builder generates 15–20 search strategies covering different angles and query shapes. More surface area means fewer blind spots in the final process.

$ research-builder generate

--task "headcount-tracking" --count 20

03

Test across company tiers.

Run the candidates against enterprises, mid-market, and SMBs. A pattern that only works for large companies is not a pattern — it's a special case.

$ research-builder test

--task "headcount-tracking" --companies tier-test.csv

04

Export the portable process.

Validated patterns graduate to a markdown file any agent can read. Drop it into Clay, Claude Code, or a custom GPT — same instructions, same results.

$ research-builder export

--task "headcount-tracking"

--output processes/headcount-tracking.md

Real scenarios

What teams actually use it for.

Not theoretical. These are the pipelines running at LeadGrow and client stacks today.

01Running consistent, high-accuracy company research inside Clay columns

02Dropping process files into Claude Code for on-demand intel pulls

03Building new research processes for tech stack, pricing, or market sizing

Included skills

Pre-built automations. Ready to run.

These aren't demos. They're the Claude Code skills we run inside LeadGrow, shipped open source so you don't have to build them yourself.

process-generator

Process Generator

Takes a research goal and generates 15–20 candidate search patterns. Scores them on specificity, reliability, and signal-to-noise. Returns a ranked list ready for testing.

Triggers: "build a research process for [goal]"

pattern-tester

Pattern Tester

Runs candidate patterns against companies across three size tiers. Scores quality and consistency. Surfaces which patterns generalize and which only work in special cases.

Triggers: After process-generator produces candidates

process-deployer

Process Deployer

Takes validated patterns and writes a portable markdown process file. Adds kill-list exclusions to cut wasted searches. Exports to the process library for reuse.

Triggers: After pattern-tester clears the accuracy threshold

More in this category

LinkedIn Monitoring

Track LinkedIn activity signals and surface warm prospects automatically.

Custom build

Want us to build yours?

Custom CLIs, MCP servers, and Claude Code skills wired to your exact GTM stack.

We build these for clients. Then we ship them open.

Book a Strategy Call →