Point it at your URL
Run one slash command with your product's URL. A 2-minute intake researches your site and confirms a few founder-only details.
Free and open-source · Apache-2.0 · Updated June 2026
A novel marketing strategy to beat competitors you can't outspend.
A free, open-source plugin that runs Diffmode's growth pipeline inside your AI coding agent — one command, about 90 minutes, and 7–9 novel tactics calibrated to your stage, budget, and channels.
claude plugin marketplace add acogood/diffmode_freeFree and open-source. Runs on your own machine — in Claude Code and OpenAI Codex.
From a URL to 7–9 tactics, in four moves
Generic AI tools stop at retrieval — a list of channels you already knew. This runs the synthesis step instead: it combines documented mechanisms into tactics that weren't in any single playbook.
It researches your market
Enrichment maps your competitors, your audience's jobs-to-be-done, and the acquisition tactics already working in your space.
Three think-tanks + a fresh vector DB
Competitor-gaps, cross-industry, and platform-arbitrage analyses run in parallel while it mines a per-run database of growth mechanisms from public case studies.
Synthesis writes your tactics
It combines 2–3 mechanisms at a time into 7–9 demand-gen tactics — about half intentionally unconventional — and writes them to synthesis.md.
Install on Claude Code or Codex
Claude Code
claude plugin marketplace add acogood/diffmode_freeclaude plugin install diffmode-growth-tactics@diffmode-free/diffmode-growth-tactics:start your-site.comOpenAI Codex
python3 codex/orchestrate.py --url https://your-product.compython3 codex/orchestrate.py --url https://your-product.com --fast-intake
Same skills, same synthesis.md. Perplexity optional — falls back to Codex's native web search.
Full setup →
CODEX.md
.
That's the whole workflow — one command, then walk away for about 90 minutes.
Before your first run: on Claude
Code, restart after installing; on Codex, make the worker agents
discoverable (see CODEX.md). A Perplexity MCP server is optional on both — connect one for the highest-quality research, or skip it and
the research stages fall back to the runtime's built-in web search (zero
setup, no API cost). The run writes into ./<slug>/ wherever you start it.
You don't need any of these to run it — the steps above are the whole workflow. For power users:
Advanced: all commands & flags (optional)
start The full pipeline — intake through synthesis. Pass your URL (or a product name) and walk away.
run-enrichment --product <slug> Enrichment only — competitors, audience, and acquisition tactics, without the think-tanks or synthesis.
| Flag | What it does |
|---|---|
--fast | Claude Code's start, URL mode: skip the ~2-minute founder Q&A and accept the researched prefill as-is — lower quality, handy for demos. On Codex, the same thing is --fast-intake. |
--url <site> | Codex's orchestrate.py: research a website to prefill the intake; the workspace slug is derived from the host. On Claude Code you pass the URL positionally instead — /diffmode-growth-tactics:start your-site.com. |
--product <slug> | Name the workspace at ./<slug>/ in the current directory. Required for run-enrichment and for Codex when you don't pass --url. On Claude Code's start, pass the name positionally instead. |
Stopped partway? Just run /diffmode-growth-tactics:start again — it auto-resumes from the last completed stage
(diagnostics,
enrichment,
think-tanks,
growth-factors,
lite-constraints,
synthesis), or choose
Start fresh to re-run from scratch. No flags needed.
A workspace you can read, not a black box
Every run writes a ./<slug>/ folder in your current directory. The deliverable is the file at the
bottom — but every file above it shows the reasoning that produced it.
-
synthesis.md — the deliverable: 7–9 novel demand-gen tactics, with about half intentionally unconventional. Each one names the mechanisms it combines and how to run it.
-
Enrichment analyses — your competitors, audience jobs-to-be-done, and the acquisition tactics already working in your space, each as its own markdown file.
-
Three think-tank reports — competitor-gaps, cross-industry transfers, and platform-arbitrage, written before any tactic is named.
-
growth-factors.json — the per-run vector database the pipeline mined from public case studies, so you can trace where each mechanism came from.
What's free, and what isn't
The method is the free part. This plugin runs the real Diffmode
synthesis, but on a deliberately lighter
per-run LIGHT database mined fresh
from public case studies every run — and it stops at ideas, your
synthesis.md.
What stays paid is the breadth and the finish: the proprietary
576-mechanism database, plus the
downstream prioritization and week-by-week implementation. The free
output is genuinely useful — full-depth ideas — and visibly lighter than
the paid product. Nothing here reads or ships the proprietary database.
What you need to run it
Claude Code or OpenAI Codex
Install the plugin into Claude Code (marketplace) or run it on Codex
(python3 codex/orchestrate.py) — same repo, same output. You spend your own Claude Code / Codex
usage per run.
Perplexity MCP (optional)
Connect one for the best research, or skip it — the research stages fall back to your runtime's built-in web search at no API cost.
About 75–90 minutes
A full run is mostly unattended. Stopped partway? Re-running auto-resumes from the last completed stage and finishes in minutes.
Roughly $0–3 per run
About $2–3 of Perplexity if you use it; $0 in API cost on the built-in web search fallback. The plugin itself is free.
Two runtimes, one method
Diffmode runs in both Claude Code
and OpenAI Codex today, off the
same GitHub repo. The skill bodies that hold the method are
runtime-neutral and live in one
place — Codex reads the exact same files through symlinks, so there's no
second copy to drift. Only the orchestration differs: a slash command in
Claude Code, a small Python driver (python3 codex/orchestrate.py) in Codex. Same inputs, same synthesis.md.
Frequently asked questions
Is it really free?
Yes. The plugin is open-source under Apache-2.0 and costs nothing to install or run. You bring your own Claude Code or Codex, and optionally a Perplexity MCP server for higher-quality research. The growth method is the free part; the proprietary 576-mechanism database and the paid implementation stages are not.
What does a run cost?
The plugin is free. If you connect a Perplexity MCP server, a full run makes about 5 deep-research and 50 search calls — roughly $2–3 depending on your plan. Without Perplexity, the research stages fall back to the runtime's built-in web search at no API cost. Either way, you also spend your own Claude Code or Codex usage.
Do I need Perplexity?
No — Perplexity is optional. Connect a Perplexity MCP server and the research stages use it for the highest-quality sourcing; skip it and they fall back to your runtime's built-in web search, with citations auto-verified for reachability. A measured no-Perplexity run matched the quality bars and finished a touch faster.
How long does a run take?
About 75–90 minutes end to end, mostly unattended. Stages overlap: the three think-tanks run in parallel and the per-run vector database is mined off the critical path. If a run stops partway, re-running it auto-resumes from the last completed stage in minutes, since the earlier outputs are cached.
How is this different from paid Diffmode?
Same method, lighter inputs, and it stops earlier. The free plugin mines a per-run LIGHT database from public case studies and ends at ideas — your synthesis.md. Paid Diffmode runs the same synthesis over the proprietary 576-mechanism database and continues into prioritization and week-by-week implementation.
Does it touch your proprietary database?
No. The plugin never reads or ships the 576-vector database. It builds its own per-run database fresh from public case studies every time, in a clean room. That is by design — the breadth of the proprietary database is part of what the paid product is.
What's the output?
A ./<slug>/ workspace in your current directory. The deliverable is synthesis.md — 7–9 novel demand-gen tactics, about half of them intentionally unconventional. Alongside it you get the enrichment analyses, the three think-tank reports, and the growth-factors.json database the run built, so you can read the reasoning behind every tactic.
Can I run it for several products or clients?
Yes. Each run names its own ./<slug>/ workspace — point it at a different product (its URL, or a name) and you can keep many side by side. Re-running an existing workspace auto-resumes from the last completed stage, so you can resume or pick up where you left off without paying for the whole pipeline again.
Does it work on OpenAI Codex?
Yes, today. Codex installs from the same repo and runs in one command — python3 codex/orchestrate.py --url … — Perplexity-optional with a native web-search fallback. The skill bodies are shared single-source (symlinks, no drift), so both runtimes produce the same synthesis.md.
Add the plugin and run your first pipeline today.
One command to install, one to run. Free and open-source — you only spend your own Claude Code or Codex usage.
claude plugin marketplace add acogood/diffmode_free
On Codex? python3 codex/orchestrate.py --url …