Growth tactics for startups that can't outspend their competitors. It researches your market, mines public case studies for the reasons real growth plays worked, combines them in ways nobody's tried, and kills anything an average marketer would have suggested anyway.
Free. No account. Runs inside Claude Code or OpenAI Codex in about 90 minutes.
graph LR
A["Your product URL"] --> R["Research"]
R --> COMP["Competitors"]
R --> BUY["Buyers"]
R --> CASE["Case studies"]
CASE --> MINE["Mine mechanisms<br/>(why each worked)"]
COMP --> CONS["Constraints"]
BUY --> CONS
MINE --> DRAW["Blind draw"]
CONS --> DRAW
DRAW --> GATE{"Average marketer<br/>already suggests it?"}
GATE -- "Yes" --> KILL["✗ Killed"]
GATE -- "No" --> OUT["7 to 9 tactics<br/>they won't see coming"]
style KILL fill:#fee,stroke:#c00
style OUT fill:#efe,stroke:#0a0
- A read on your competition. Who you're up against, how each rival gets users, and which channels they're ignoring. That gap is where you get in.
- A map of your buyers. Real segments, the job each one hires you to do, and what makes someone switch.
- 7 to 9 unconventional tactics. Each one a plain card: what it is, why your competitors won't see it coming, how to start this week, and when to kill it.
Here's the kind of tactic a run produces:
The Hiring-Signal Pitch. When a company posts a job to hire someone for the exact manual task your product removes, that's a buying signal nobody else is watching. Set alerts for those job titles. When one's posted, find the hiring manager and send a 60-second screen recording: "saw you're hiring a [role] to do [task]. Here's my tool doing it live." A reply rate above ~10% means the angle lands.
Everything is yours to reuse. Hand the competitor read to a freelancer, drop the buyer map into a deck. The run ends with a styled report in your browser.
Most marketing tools hand you a checklist: do SEO, run ads, post on LinkedIn. Diffmode does the opposite.
Mechanism, not tactic. It stores why growth plays worked, not which channels were used. Mechanisms transfer across industries; channel checklists don't.
Blind before analysis. It picks which mechanisms to combine before it knows what tactic they'll produce. That stops the model from reverse-engineering its way back to the obvious answer.
Rejection is the product. Four separate gates throw away anything an average marketer would recommend. Novelty isn't generated. It's what survives.
Claude Code (two commands, then restart):
claude plugin marketplace add acogood/diffmode_free
claude plugin install diffmode-growth-tactics@diffmode-freeThen from any folder:
/diffmode-growth-tactics:start your-product.com
No flags. Run it bare and it asks for your website. Run it again in the same folder and it resumes where it stopped.
Codex (clone this repo, then run):
python3 codex/orchestrate.py --url https://your-product.comSame pipeline, same quality gates. Full setup in codex/CODEX.md.
- Claude Code or OpenAI Codex.
- Research backend: the built-in web search, free, nothing to set up. A Perplexity MCP also works if you already have one (about $2-3 per run). Both are fully supported.
- Model: synthesis runs on Opus for reasoning quality; research and packaging use Sonnet.
This free tool builds the strategy: the competitor read, the buyer map, and the unconventional ideas. It stops at ideas.
Diffmode picks up from there. It ranks the tactics so you know what to run first, and turns the top picks into a week-by-week rollout plan, drawn from a much deeper database of growth mechanisms. Start with a free audit (no credit card). The full plan comes with a 30-day money-back guarantee.
13 skill files hold the methodology. Four stateless workers execute them. An orchestrator owns the DAG, the retries, and the quality gates. Skills live once in plugin/skills/; Codex consumes them through symlinks. Full design in docs/architecture.md.
License: Apache-2.0. © 2026 Anton Kogut.