Plain markdown + JSON. If your agent reads files, it trains here.
You built something good. Marketing is why nobody knows it exists. Install the brain: your agent plans your launch, writes copy that sells, finds where your buyers hang out and runs distribution every single week. You approve. It executes.
Plain markdown + JSON. If your agent reads files, it trains here.
Ask your agent for a landing page or a thread and you get copy that smells like every other AI product on the internet. Because it is working from the same training data.
You ship all day, then do positioning, hooks and distribution at night, badly. The thing you are worst at decides whether anyone sees the thing you are best at.
Free marketing skills on GitHub are written once and frozen. Your agent confidently executes 2024 playbooks in a market that moved three times since.
One prompt, run twice on the same agent. The only change between the two runs: Agent Gym installed.
Run npx agent-gym init in your repo. The full pack lands as files your agent reads natively.
Answer 5 questions about your product, niche and ICP. Every skill, hook and drop is tuned to them from day one.
Every Monday at 07:00 a fresh signal drop lands. Your agent ingests it, levels up, and works from this week's market, not its training data.
agent-gym/
├── skills/ 27 skills across 6 modules
│ ├── strategy-core/ positioning, ICP, offer, funnel
│ ├── content-engine/ landing, threads, email, scripts
│ └── ...
├── vault/
│ ├── hooks.json 596 entries, scored + tagged
│ ├── formulas.md 41 working patterns
│ └── swipes/ winning ads, refreshed monthly
├── signals/
│ └── drop-w27.md fresh every Monday 07:00
└── profile/
└── icp.md tuned to your product in step 2
No dashboard, no login, no 6-hour video program. You keep prompting your agent the way you already do. It just answers like a senior marketer now, and cites its sources.
what are our top content opportunities this week?
Found 5 high-impact plays from this week's drop, ranked by search demand + competitor gaps in your niche.
turn #1 into a launch thread
Using hook #217 (outperforming 4:1 this week) + your ICP profile. Done: 9 posts, CTA to trial signup, seeded comment included. Best posting window for your niche: Tuesday 09:40.
We tested 214 skills on live projects. 27 made the cut. Anything that stops performing gets replaced, and the update syncs to you.
The layer most founders skip, and the reason their content converts nothing.
Each skill ships with worked examples, not blank templates.
Agent-readable databases of what already worked, in markdown and JSON.
The weekly drop your agent ingests on its own. It stops guessing from 2024 data.
New pages, ads and angles from competitors you name, turned into action.
Marketing that reviews itself. Your agent learns from its own shipped work.
The 5-question setup tunes every skill to your stage, not just your niche. Here is what day one looks like.
I ship this month. Nobody is waiting for it.
Good product. Twelve visitors a day.
$5k MRR. Zero time for marketing.
While you code, we scan what actually converts right now: which formats blow up, which hooks die, what your competitors ship. Your agent gets the digest and applies it to your product. Nobody else sells this part.
Posts crossing engagement thresholds in AI, SaaS and builder niches, decomposed into reusable hook patterns.
Founder subreddits, Indie Hackers and niche Discords: raw pain language and questions people actually ask.
Meta and TikTok ad libraries: creatives running 30+ days are creatives that convert. They enter the swipe file.
Keyword volume spikes and which products ChatGPT and Perplexity cite this week, per category.
Product Hunt and launch-week teardowns: which positioning and pricing shipped, what stuck, with numbers.
Algorithm and policy shifts on X, LinkedIn, TikTok and YouTube, translated into revised posting patterns.
Weekly scrape of competitors you name: new pages, ads and angles, delivered as counter-briefs.
We run these exact skills on real client systems daily. What underperforms gets replaced before it reaches you.
sources monitored weekly
raw signals collected
survive relevance + freshness filters
make the Monday drop, tuned to your niche
Plain markdown. Your agent reads it on the next session, no dashboard, no copy-paste.
New hook patterns reweight the Vault, stale advice gets flagged, the niche profile absorbs this week's trends. +50 xp
Every drop ships one concrete play: a pricing-page loop, a trend post, a counter-brief. Your agent drafts it, you review and ship. +150 xp
The Analytics Loop scores what you shipped against the databases. Next Monday's drop is tuned by what actually performed for you. level up
A real drop file, filtered for your niche. No dashboard, no login. It lands in your repo and your agent takes it from there.
"Agent-readable docs" search volume +312% in 90 days
angle: publish an llms.txt teardown before your competitors do
"I deleted [tool] and replaced it with one prompt" outperforming listicles 4:1
Two rivals shipped comparison pages targeting your keyword
counter-play included: 340-word response brief
X now throttles link-in-first-reply posts too
revised posting pattern included, tested on 3 accounts
Run the new pricing-page loop on your own product. +150 xp
or $99/year (2 months free) · cancel anytime
One install. The output of a marketing hire. The price of a tool.
Get the full stack ↗ Start with the free tier first if you want. No card required."Dropped the pack into Claude Code, ran the onboarding prompt, and by Friday my agent wrote a landing hook that beat the one I paid a $400 copywriter for."
"The Monday drop replaced two hours of my own trend digging. My agent now pitches angles I would not have found myself."
"Bought it for the skills, stayed for the signal drop. My agent is level 9 and my content calendar runs itself."
| Marketing hire | Freelancer | Free skill repo | Agent Gym | |
|---|---|---|---|---|
| Cost | $6,000+/mo | $2,000+/mo | $0 | $15/mo |
| Knows your niche | yes | partly | no | tuned at onboarding |
| Fresh market data | sometimes | sometimes | frozen at install | new drop every Monday |
| Works nights + weekends | no | no | n/a | it's an agent |
| Gets better over time | maybe | no | no | levels up weekly |
The founders who get this early are shipping product and distribution from the same terminal.
Free
$0
try the gym
Core
$15/mo
or $99/year, 2 months free
Pro
$49/mo
for operators in a niche
Founding
$299 once
Pro, forever
37 of 50 spots left
Use it for 30 days. If your agent's marketing output is not noticeably better, one email gets you a full refund. No forms, no calls, no "exit survey". We keep the risk, you keep the upside.
A folder your agent reads: 27 skill files across 6 modules, the Vault (596 hooks, 41 formulas, CTA scripts, ad swipes), 9 distribution playbooks, the 3 bonuses, an onboarding prompt that personalizes everything to your product, and a new signal drop file every Monday. Plus the leveling system and your agent.card.
Yes. Everything ships as plain markdown and JSON. Any agent that can read files in your repo can train here: Claude Code, Cursor, Codex CLI, Gemini CLI, Copilot Workspace and most custom setups.
Curation, freshness and personalization are the product, not the files. Free repos freeze at install. Your subscription pays for weekly re-testing on live projects, replacement of anything that slips, and a drop that keeps your agent on this week's market.
No. The skills carry the strategy: your agent asks you about the product, then handles positioning, copy and distribution the way a senior would. You review and ship. You will accidentally learn marketing by reading its output.
No. Agent Gym runs on the AI subscription you already pay for. We never proxy your API calls, so there is no markup and your code never touches our servers.
The signal drop lands every Monday at 07:00. Skills and the Vault are re-tested weekly on live projects; anything underperforming gets replaced, and updates sync to you automatically.
Setup is 10 minutes. The first noticeable difference is usually the first thing your agent writes afterwards, because it cites the Vault instead of inventing generic copy. Compounding kicks in from the first Monday drop. The guarantee gives you 30 days to judge.
30 days, full refund, one email, no questions. If your agent's marketing output is not noticeably better, we do not want your money.
Install the free tier tonight. The first drop lands Monday at 07:00.
Start free ↗