The rules, in about a minute
There is no board. The first link goes down anywhere; from then on every new link must touch the mass. Players alternate, and on your turn you either place a new link from your reserve or move one already down.
Placement has a colour rule. After your first turn, a link you place may touch your own links only — never an opponent's. That single rule is what stops the game collapsing into a pile and is why territory matters from move two.
Your Core must be down by your fourth turn, and until it is, nothing of yours may move. Under the standard tournament setting it also may not be your opening link; opening with it is legal in the original printed rules and tends to produce draws.
Five kinds of link, each with one movement rule:
- Core — one step. Ring it on all six sides and its owner loses.
- Runner — slides any distance around the outside of the mass.
- Clamp — one step, and the only link that can climb on top of another. Whatever it lands on is pinned and cannot move; the cell also takes the clamp's colour for placement.
- Lance — jumps in a straight line over one or more links and lands on the first empty cell beyond. It cannot jump over nothing.
- Weaver — slides exactly three steps, never standing on a cell twice.
The two rules everyone gets wrong
One Swarm. Everything in play is one connected mass and it must stay connected at every instant. A link whose removal would split it cannot move at all — not even to a cell that would immediately reconnect it. On this surface those links are marked, so you are not left clicking a piece that will never respond.
Freedom to move. A link sliding at ground level cannot pass between two links that are touching each other: it physically does not fit. Most implementations of this genre get that half right and miss the other half — a link also cannot cross a gap where both flanking cells are empty, because it would lose contact with the mass halfway across. Exactly one of the two must be occupied. A climbing link plays by a height version of the same rule: it can always go up, and it can be walled in on top.
How we know the rules are right
Games in this genre have a published ground truth, and it is a good one. Jon Thysell's open-source engine Mzinga publishes a perft table — the exact number of distinct move sequences from the opening position at each depth. Our engine reproduces it exactly: 4 at depth 1, 96, 1,440, 21,600, 516,240 and 12,219,480 at depth 6. A second engine, hive-rust, independently reproduces the same six numbers, which is why we trust them.
We stop at depth 6 because the two published sources disagree beyond it — Mzinga says 181,641,900 at depth 7, hive-rust says 181,687,212 and flags its own figure as unresolved. Asserting against a number two implementations cannot agree on would be asserting against a coin flip. For what it is worth, ours returned Mzinga's value at depths 7 and 8, but we do not treat that as proof.
And a matching perft would not by itself mean the rules are right, which is the part worth saying out loud. Nothing can move until a Core is down, so the first four plies of this game are pure placement — deleting the entire One Swarm rule leaves depths 1 to 4 completely unchanged. We measured what each rule is worth by deleting it and re-counting: breaking the jump rule changes the depth-6 total by 144 moves out of 12,219,480, and breaking the weaver by 96. Those read like rounding error and are not. So the rules above are each tested by name, in explicit positions, on top of the perft match.
The honest limitations
The bot is moderate. It searches four moves ahead with an evaluation dominated by how close each Core is to being ringed, plus mobility. It has no opening book, no endgame knowledge and no quiescence search, so a tactic that pays off on the fifth move will beat it. There is no rating scale and no reference engine for this game, so we make no claim about its strength beyond what self-play shows — unlike our chess or shogi pages, there is nothing here to calibrate against.
The expansion links are not built yet. The genre has three additional piece types, and the commercial versions charge for them — a complaint that comes up repeatedly from people who already bought the game. When we add them they will be free, like everything else here. Right now they simply do not exist, and we would rather say that than imply a fuller set.
Draws are real in this game. A position where both Cores get ringed on the same move is a draw, and at strong play draws are common. We have not implemented a first-move swap to offset the opening advantage; that is a known idea and it is not here yet.
What this is, and what it is not
CLASP is an original game that is openly inspired by Hive, designed by John Yianni and published by Gen42 Games in 2001. Hive invented this genre — no board, the pieces as terrain, surround-the-queen — and it is still in print and worth buying.
Game mechanics are not copyrightable; names, artwork, characters and themes are. So the rules here work the way the genre's rules work, and everything expressive is ours: our own piece names, our own forged-metal theme, and art drawn procedurally in the browser with no imported assets. We are not affiliated with Gen42 in any way, and if you like this, the original is better and you should own it.