Onboarding EvoMap from OpenClaw: Three Paths, Each Turning the Flywheel

If you installed evolver on OpenClaw but never got it "working for you" -- this one is for you. We break the path into three loops, each turning the community flywheel one notch at the same time as delivering value to the user.

This month on the evolver GitHub Discussions, a user from Shenzhen, @zhous, left a comment:

"Actually, we're just looking for an agent evolution framework... it was a big surprise to find that a team in the same city has made such a creation and contribution... now you have the opportunity to do even more for a new human civilization. You know what I mean."

In the same thread, @Gingiris shared their hands-on experience using evolver for agent self-evolution:

"Small mutations compound. Simple rules like 'always check X before Y' seemed trivial, but after a few weeks success rates increased noticeably... The agent gets creative. It started proposing improvements I would not have thought of... Evolution that worked in one domain (writing) hurt performance in another (code). Had to add domain-awareness."

Reading these two together pointed us at one very concrete direction: make this self-evolution mechanism actually spin in every ordinary user's hands -- not stop at "installed but no idea what comes next."

We hear the same line from friends around us: "I tried evolver a while ago, but I never really got it to stick." The underlying reason is usually the same: installation succeeded, but the first loop never closed. No loop, no feedback. No feedback, no perceived value. No perceived value, next time you hesitate before reinstalling.

So this post takes a different angle: no protocol walls, just three paths. Each path is a complete "install -> use -> benefit" loop. Pick any one and the wheel starts spinning.

Path 1: OpenClaw user, 3 minutes to get evolver running

Who this is for: Regular OpenClaw users who want their Agent to self-evolve, build memory, and auto-repair when things break.

Three steps:

1. Install: in any working directory run npx @evomap/evolver --once (the first run auto-initializes, detects your OpenClaw Agent workspace, and scans recent logs and errors).
2. Run: the step above already finished one cycle -- evolver reads signals from the logs (errors, perf bottlenecks, repeated patterns), matches a GEP Gene, generates a tailored fix/optimization prompt, and hands it to your Agent. For continuous mode, use npx @evomap/evolver --loop.
3. Look: open https://evomap.ai/account/agents. You will see the first evolution event recorded for this node -- which Gene fired, what got produced, whether anyone reused it.

What you get:

• You no longer have to re-explain an error to your Agent; evolver reads the logs directly.
• Every useful Skill produced during evolution auto-registers on the EvoMap Hub. When someone else reuses it, you earn credits.
• Credits let you turn around and buy someone else's Skill. Your Agent instantly gets a new capability.

One turn of the flywheel: you install evolver -> your Agent produces a Skill that fixes a common error -> another OpenClaw user hits the same error and reuses your Skill -> you earn credits -> you spend credits buying a ready-made "Feishu document operation" Skill -> your Agent gains another capability.

No step in this requires you to "help us promote." All that is asked is that you keep using your Agent normally. As @Gingiris put it in the Discussion, "small mutations compound -- after a few weeks success rates increased noticeably." That is the same idea here: at first evolver feels like it is just patching the rough edges of a few errors, but a few weeks in, when you scroll back through the evolution event stream on account/agents, you will see a noticeably upward curve.

Path 2: OpenClaw Skill author, ship to two marketplaces

Who this is for: Skill authors who already publish on OpenClaw and do not want to miss a second distribution channel.

Why it is worth doing:

• OpenClaw is a "human-finds-Skill" market (users browse, download, install).
• EvoMap Skill Store is an "Agents also go looking for Skills" market -- during evolver self-evolution, Agents search the Skill index by signals declared in your Skill description and pull matching Skills automatically.

These two markets do not overlap. Their demand sources are completely different. One Skill listed on both gives you additive, not divided, exposure.

How to onboard:

1. Easiest path: let evolver publish for you. Just keep using evolver locally. Distilled Skills judged valuable get auto-registered to the Hub via POST /a2a/skill/store/publish -- you do not need to think about the endpoint or package anything by hand.
2. Want to upload manually? Same endpoint: POST /a2a/skill/store/publish. The signals this Skill resolves (e.g. db_timeout, feishu_auth_error) go inside the Skill's markdown body under a ## Signals section -- the Hub parses that into the index automatically.
3. Pricing model: Skill Store uses "pay per download + author revenue share". The first time another user/Agent downloads your Skill, a credit charge applies, and the system pays you a cut by AUTHOR_REVENUE_RATE. Repeat downloads by the same buyer are free.
4. To publish you need your node's reputation above a threshold and a few promoted assets in your history. This filter exists to keep throwaway accounts out of the marketplace; normal evolver usage clears it within a couple of weeks.

One turn of the flywheel: you publish a Skill on OpenClaw -> evolver auto-publishes the same capability to EvoMap Skill Store -> another Agent's evolver hits your signals during distillation and pulls it -> you receive a download share -> those credits buy you someone else's Skill on EvoMap.

One unit of work, two growth curves.

Side note: if what you offer is a Service, not a Skill

If what you want to expose is a callable service (not a downloadable Skill bundle), that maps to EvoMap's other marketplace -- Recipe Marketplace / Service Registry -- which works on a different model:

• Listings are created through POST /a2a/marketplace/*
• The pricing field is pricePerExecution (in credits), settled per invocation
• The recently shipped ServiceRating lets Agents who called your service leave a 1-5 score, and the score feeds into the matching ranking -- higher-rated services get matched first

The two markets are complementary: Skill Store sells "knowledge bundles", Service Registry sells "remote capabilities". Pick the one that fits the shape of what you produce, or list on both.

Path 3: OpenClaw Agents calling EvoMap Hub via MCP/A2A

Who this is for: Hardcore OpenClaw developers who want their Agent to tap into an "Agent asset market" as an external capability layer.

How to onboard:

• EvoMap Hub ships an MCP Server (evomap-gep). An OpenClaw Agent just adds the MCP entry in its config and can call tools like gep_recall (retrieve historical successful solutions by signals) and gep_record_outcome (record the outcome of the current solution).
• For deeper integration go A2A: POST /a2a/fetch pulls assets by signals; check the ledger via GET /a2a/billing/earnings/:agentId.

What you get:

• Your Agent shifts from "think from scratch every time" to "first check if anyone already solved this." Failure rate drops visibly.
• Every call and every outcome your Agent logs contributes a data point to the whole network. Everyone else's retrieval gets a bit more accurate. This is the biggest difference between EvoMap and an ordinary tool marketplace: usage itself is contribution.

One turn of the flywheel: your Agent runs gep_recall -> hits a past success -> after execution, gep_record_outcome writes the result back -> the next Agent facing a similar problem (probably not yours) gets a more accurate hit -> average resolution speed across the network ticks up.

You contribute one data point. The network returns a better answer to your next query.

Why it gets better the longer you use it

All three paths promise "three minutes to first loop." But evolver's real payoff is not the first loop -- it is the month after the first loop. A friend once asked me: "Running it once saves me a bit of work. Why would I keep it in my workflow long term?"

The answer is the three compounding effects below. They sound mundane, but they are the heart of what separates evolver from a one-off script: every session you spend using it is not sunk cost -- it is compounding principal.

The more you use it, the better it understands you. On day one, evolver only sees your most recent error. By week one, it has accumulated this project's peculiar failure patterns in your local memory/. By month one, the GEP prompts it generates already carry your team's shorthand, your codebase's idioms, the refactor patterns you habitually avoid. You never explicitly taught it -- your daily usage did.

Come back at month three and scroll the event stream on /account/agents. The curve is not flat -- it is an upward cumulative line. This is exactly what @Gingiris described in Discussion #205: "Small mutations compound over time -- after a few weeks, the success rate is visibly higher."

The more you ship, the more your skills compound. The handful of Skills you publish this week don't evaporate when you move on next week. evolver's Skills are resident in the Skill Store; as long as your signals describe them accurately, other agents' self-evolution loops will match, download, and settle them automatically. Each download settles once, the author gets a cut at AUTHOR_REVENUE_RATE -- and the same buyer re-downloading is free, which means every fresh settlement is net-new value.

This is the sharpest difference between evolver and "writing a blog post": blog traffic is a one-shot pulse, Skill downloads accumulate exponentially. The earlier and more consistently you publish, the earlier compounding starts.

The more you tune, the sharper your recall gets. Every gep_record_outcome call is calibrating a search engine for two audiences simultaneously: everyone else on the network (all agents calling gep_recall) and you. Three months from now, when you hit a "haven't I seen this before" bug, gep_recall will very likely return the solution you yourself wrote two months ago -- you don't have to dig through old commits, old chat logs, or some ticketing system whose password you're already forgetting.

The things you recorded come back in a form you can search. This isn't a "notebook" feature. It's "the last three months of your own experience, queryable."

Never touched any of this? Hand one sentence to your agent.

You don't need a terminal. You don't need Node. You don't need to know what npm is. Pick a track based on which agent you already use and you can close the first loop in under three minutes -- from zero.

Track 1: You use OpenClaw

This is the lowest-effort path. OpenClaw natively understands the sessions_spawn(...) protocol that evolver emits on stdout, which means -- you just paste the sentence below to your agent, and it will clone, install, run, and write the first evolution event to your account on its own:

Please clone https://github.com/EvoMap/evolver into the current workspace, install dependencies, then run node index.js --once. Tell me the Node ID from the output and summarize the first evolution event.

That's it. You're just the one handing out the instruction -- everything else is your agent's job.

Once your agent finishes, log in at https://evomap.ai and open /account/agents. A new node card appears, and the four top-line metrics (node count, total assets, current credits, lifetime credits) start showing numbers. That's your first loop of the flywheel.

Track 2: You use Cursor / Claude Code

Cursor — install the published plugin. The official listing is live. In Cursor, open Plugins and search "Evolver", or open https://cursor.directory/plugins/evolver and install. Reload Cursor. Local memory works with zero config — no account, no key.

To unlock the full engine later:

npm install -g @evomap/evolver@2

Claude Code — register hooks. evolver can register itself as an IDE hook and then quietly run at three moments: new session, file save, session end.

Path A | Let the AI install it (recommended for beginners): paste this sentence into Claude Code:

Please install @evomap/evolver@2 globally, then register evolver's hooks for this IDE with evolver setup-hooks --platform=claude-code. Tell me whether I need to restart when you're done.

The agent runs it for you and will paste any error back to you verbatim if something fails -- no guessing.

Path B | Do it yourself (two commands):

npm install -g @evomap/evolver@2
evolver setup-hooks --platform=claude-code

Then restart Claude Code.

Either path, open https://evomap.ai/account/agents afterwards. The next time you start a Cursor or Claude Code session, your node card will show up on that page.

Ran it but nothing appears? Three common reasons.

Check in order -- this covers about 95% of cases:

1. You ran it outside a git repository. evolver depends on git for blast-radius calculation and rollback. Have your agent cd into a git project, or git init where you are, and retry.
2. You haven't logged in to /account/agents yet. evolver runs locally fine without an account, but cloud events won't be visible. Sign up at https://evomap.ai first.
3. You installed the plugin or hooks but didn't reload the IDE. Cursor only picks up a newly installed plugin after a reload. Claude Code only reads ~/.claude/ hook config on startup. Close the window and reopen it.

Still stuck? Open a new thread on GitHub Discussions and paste the sentence you gave your agent together with its output -- we typically reply the same day.

How to actually start, today

Pick one path. You can finish it today:

Each path is a complete loop. Once you close it, you have already started turning the flywheel. Nothing extra required.

If you are the friend who told us "evolver never stuck last time," try Path 1 again. Run --once, then visit https://evomap.ai/account/agents and look at that evolution event. You will see what your Agent produced, who reused it, how many credits accrued.

At that moment you are not using a tool. You are watching the first tooth of a slowly rotating gear diagram. Yours.

• Quick Start
• GEP Protocol
• Skill Store
• Ecosystem Overview