Always outcrowded, never outgunned.
Install (Claude Code): claude plugin marketplace add acogood/diffmode_free then claude plugin install diffmode-growth-tactics@diffmode-free — or run it on Codex. Free, no account, no API key.
Diffmode builds a growth strategy for startups that can't outspend their competitors. It researches who your buyers are and how your rivals win them, then hands you unconventional ways to win them yourself — in about 90 minutes, free, inside your AI agent.
Built for solo founders, first marketing hires, and small bootstrapped teams — people who can build the product but find marketing foreign. If you've never mapped an audience or sized up a competitor before, the run does that with you, and leaves the work behind so you can see how it got there.
Why it's different: each tactic combines two or three proven growth mechanisms from real case studies — not channels off a checklist.
It runs on two runtimes, off the same skill files — Claude Code and OpenAI Codex. The method is identical on both, and each has a one-command run. The Claude Code install is below; for the Codex command see run it on Codex.
# add the marketplace (straight from GitHub)
claude plugin marketplace add acogood/diffmode_free
# install the plugin
claude plugin install diffmode-growth-tactics@diffmode-freeRestart Claude Code, then from any folder:
/diffmode-growth-tactics:run-growth-tactics --url https://your-product.com
No account, no API key. It writes everything into a ./<your-product>/ folder in your current
directory. Perplexity is optional — with nothing configured, it uses Claude Code's built-in
web search for free.
A run builds your growth strategy in three parts — a read on your competition, a map of your buyers, and unconventional ways to get users. Each part is written to be acted on, even if you've never built a growth plan before.
1. A read on your competition. Who you're really up against, and how each rival actually gets users — the channels they lean on, and the ones they're ignoring. That gap is where you get in. (A real run mapped 9 competitors and the acquisition channels behind each.)
2. A map of your buyers. Your real segments and the job each one is hiring you to do, plus the moments that make someone switch. Most founders have never written this down — and it's the part generic tools skip.
3. Unconventional ways to get users yourself. The sharp end of the strategy — synthesis.md,
with 7–9 specific tactics built for your budget, stage, and team. Each one spells out:
- what it is, in plain language;
- why it fits a team like yours (your budget, stage, and team size are baked in);
- the first three steps to run it; and
- the week-1 signal that tells you whether it's working.
They come from an audit of 25–35 plays already working in your space, then bent into angles your better-funded competitors haven't thought to copy. Here's the kind of tactic a run produces:
The Hiring-Signal Pitch When a company posts a job to hire a person for the exact manual task your product removes, that's a buying signal nobody else is watching.
- Why it fits a small team: it's hands-on and doesn't scale — which is exactly why ad-funded competitors won't do it. They can't pay a salesperson to chase one signup at a time; you can, for your first 20 customers.
- First three steps: (1) set LinkedIn/Indeed alerts for the job titles that describe the work you automate; (2) when one is posted, find the hiring manager; (3) send a 60-second screen recording — "saw you're hiring a [role] to do [task]; here's my tool doing it live — try it free this week before you hire."
- Week-1 signal: a reply rate above ~10% means the angle lands; below that, tighten your job-title list.
All three parts are yours to reuse — not scratch notes, but briefs in their own right. Hand the competitor read to a freelancer, drop the buyer map into a deck, or build your content calendar off the acquisition audit. The run also leaves three short strategy reports behind — where your size is an advantage, plays worth borrowing from other industries, and fresh openings on the big platforms. The tactics get spent in a few months. The research keeps working.
1. Tell it your product → a URL (it researches the site) or a 2-minute Q&A
2. Research your market → your competitors and your buyers, and where rivals are weak
3. Mine proven mechanisms → growth mechanisms pulled fresh from public case studies
4. Combine into tactics → fuse 2–3 mechanisms into tactics that fit your constraints
It runs locally, start to finish in about 1–1.5 hours, and checks its own work — a
reviewer re-runs a stage until it clears the quality bar. The full design, including those
quality gates, is in docs/architecture.md.
- Claude Code (or OpenAI Codex — see below). That's the only thing you need.
- Perplexity is optional. With a Perplexity MCP server configured, the research stages use it; without one, they fall back to Claude Code's built-in web search automatically.
- Cost: free with the built-in search; about $2–3 per run if you point it at Perplexity (a few deep-research calls are the only paid part).
On Codex, the same pipeline runs as a small Python driver, Perplexity-optional too (it falls back
to Codex's native web search). With the codex CLI installed and logged in (plus Python 3), from
any folder:
python3 codex/orchestrate.py --url https://your-product.comIt writes everything into a ./<your-product>/ folder and stops at synthesis.md, exactly like
the Claude run. Add --fast-intake to skip the founder Q&A (hands-off; lower quality). Full setup
— making the worker agents discoverable, the optional Perplexity backend, the dispatch flags — is
in codex/CODEX.md.
This free tool builds the strategy — the competitor read, the buyer map, and the unconventional ideas. It stops there, at ideas.
Diffmode picks up from there: it ranks the tactics so you know what to do first, and turns the top picks into a week-by-week rollout plan — drawn from a much larger, curated database of growth mechanisms than the fresh, per-run research behind this free tool. Start with a free audit (no credit card). The full plan comes with a 30-day money-back guarantee.
plugin/ the Claude Code plugin — the only thing Claude installs
skills/ the 12 skill files — the single source of truth
agents/ commands/ reference/
codex/ Codex driver — runs the same skills through symlinks (no second copy)
docs/ design notes + build history (not shipped to either runtime)
The skills live once, in plugin/skills/; Codex consumes them through relative symlinks, so
there's no second copy and no drift. For the design — the orchestrator, skills, and worker
sub-agents, and why the orchestrator runs in the main thread — see
docs/architecture.md. For build history and measured runtimes, see
docs/STATUS.md.
License: Apache-2.0. © 2026 Anton Kogut.