Say things a player can act on, and test the crash containment

[nicodes]

The polish pass from the readiness audit. Four small things, all the same complaint: the arcade explained itself to a developer.

Stacked on #8 (which is stacked on #7). Merge in order and each retargets cleanly.

Errors a player can act on

The registry client turned a transport failure into Go's error, and the TUI rendered it straight into a notice — so a marketplace that was merely not running looked like a stack trace.

before  registry unreachable: Get "http://127.0.0.1:8080/v1/games/aviorstudio/tetris/resolve?abi=1": dial tcp 127.0.0.1:8080: connect: connection refused
after   the marketplace is not answering — check your connection and try again

The detail is behind TERMCADE_DEBUG for whoever is actually debugging:

$ TERMCADE_DEBUG=1 termcade add aviorstudio/tetris
termcade: the marketplace is not answering — check your connection and try again: Get "http://127.0.0.1:59999/..." dial tcp ...: connection refused

Fixed in the client, so the CLI and the TUI both get it — they were rendering the same string. The registry's own messages now pass through unprefixed, since termcade: registry: ... was a stutter rather than attribution, and the TUI stopped prefixing marketplace unreachable: onto a message that already says so.

Empty state for list

With nothing installed it printed nothing at all and exited 0, which reads as a broken command rather than an empty arcade. The TUI has empty states on every screen; the CLI had none.

$ termcade list
no games installed — run `termcade` and press m for the marketplace

A bare slug says what's wrong with it

Forgetting the author is the likeliest way to get an id wrong, and the old answer was about the filesystem.

before  open asteroid: no such file or directory
after   "asteroid" is missing an author — marketplace ids look like author/slug, e.g. aviorstudio/asteroid

Anything with a slash, a scheme, or a .tcade on the end still means a file or a URL and is left alone.

SafeGame has tests now

It had none, which was the wrong place for the gap — it is the whole of "a broken game can never take the arcade down". Eight tests:

• every method (Reset, HandleKey, HandleKeyUp, Draw, Update, Score, HUD) contains a panic rather than propagating it
• a panic in Info, which happens inside Safe itself before the caller holds anything
• the first crash latches and nothing reaches the game afterwards — a game panicking every frame doesn't burn the arcade's time
• Close still runs after a crash, which is exactly when a wasm instance needs releasing
• a panic in Close is contained too, and a real Close error is passed through
• a healthy game is untouched

Verification

go vet clean, full suite passes. Every message above was checked against a clean install with the registry pointed at a dead port.

🤖 Generated with Claude Code