ai-sandbox-daytona: spawn().stdin.end() is a no-op while writableStdin: true, so stdin-fed harnesses (Claude Code, Codex) hang forever
I maintainer di solito rispondono entro 1 giorno
Nessuno ha ancora preso questa issue.
Valutazione
- Difficoltà
- 3/5
- Tempo stimato
- 1-2 giorni
- Idoneità per principianti
- 64/100
- Tipo di issue
- Bug
- Chiarezza
- Abbastanza chiara
- Stato di attività
- Attiva
- Stack tecnologico
- typescript
Direzione di ricerca
Start by reading the Daytona spawn handle used by spawnNdjson and startJournaledAgent in ai-sandbox, then check how the Daytona API handles closing stdin and what tests cover stdin writes. Confirm that ending stdin delivers EOF and lets stdin-fed processes finish; the issue is resolved when the reproduction no longer hangs.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
TanStack AI version
1.0.0
Framework/Library version
React 19
Describe the bug and the steps to reproduce it
What happens
On a Daytona sandbox, a Claude Code run never starts working. claude -p starts, writes its config, and then waits with no output until the run is aborted. The claude process stays alive in the sandbox afterwards, because killableProcesses is false. Codex takes the same stdin path, so it hangs the same way.
Cause
DAYTONA_CAPS.writableStdin is true, so the harnesses send the prompt through stdin. spawnNdjson and startJournaledAgent in ai-sandbox call proc.stdin.write(input) and then proc.stdin.end(). Daytona's spawn handle implements end as a no-op:
stdin: {
write: (data) =>
this.sandbox.process.sendSessionCommandInput(sessionId, cmdId, data),
end: () => Promise.resolve(),
},
sendSessionCommandInput delivers the bytes, but EOF never arrives. claude -p and codex exec read stdin to EOF before they begin, so they wait forever.
Minimal reproduction
import { daytonaSandbox } from "@tanstack/ai-sandbox-daytona";const provider = daytonaSandbox({ apiKey: process.env.DAYTONA_API_KEY });
const h = await provider.create({} as never);
const read =node -e "let s='';process.stdin.on('data',d=>s+=d).on('end',()=>console.log(s))";// 1. Through stdin: never exits.
const p = await h.process.spawn(read, {});
await p.stdin.write("hello");
await p.stdin.end();
console.log(await Promise.race([
p.wait().then((c) =>exit ${c}),
new Promise((r) => setTimeout(() => r("still waiting after 20s"), 20_000)),
])); // -> "still waiting after 20s"
// 2. Through a file redirect: exits at once.
await h.fs.write("/tmp/prompt", "hello\n");
const q = await h.process.spawn(${read} < /tmp/prompt, {});
console.log(await q.wait()); // -> 0, prints "hello"
Measured on a real Daytona sandbox (daytonaio/sandbox:0.8.0): case 1 was still waiting after 20 s, and case 2 exited 0 in about 2.4 s.
Affected harnesses
| Harness | How the prompt is passed | On Daytona |
|---|---|---|
| ai-claude-code | stdin (claude -p) | hangs |
| ai-codex | stdin | hangs |
| ai-grok-build (NDJSON) | argv (-p "") | not affected |
| ai-opencode | HTTP to opencode serve via ports.connect | not affected |
Suggested fix
Either of these would fix it:
- Set
writableStdin: falsefor Daytona until EOF can be delivered. The harnesses already have this path: they write the prompt to a file and run… < file, and it works on Daytona today. This is the smallest change. - Implement
end()for real, if the Daytona session API can close a command's stdin (for example, a PTY session, which #1085 mentions). If it cannot,writableStdin: trueshould not be advertised.
A test for whichever fix is chosen: spawn a process that reads stdin to EOF, then write and end, and assert that it exits.
Workaround
We wrap the provider so that it and its handles report writableStdin: false:
const handleWithoutStdin = (h: SandboxHandle) =>
Object.defineProperty(h, "capabilities", {
value: { ...h.capabilities, writableStdin: false },
});
// applied to provider.capabilities(), create(), resume(), restoreSnapshot()
After this, ai-claude-code takes the prompt-file path, and the prompt reaches the CLI.
Related: #1085 (session stdin, PTY kill).
Your Minimal, Reproducible Example - (Sandbox Highly Recommended)
Code Sanbox
Screenshots or Videos (Optional)
No response
Do you intend to try to help solve this bug with your own PR?
None
Terms & Code of Conduct
- I agree to follow this project's Code of Conduct
- I understand that if my bug cannot be reliable reproduced in a debuggable environment, it will probably not be fixed and this issue may even be closed.
- Lingua principale
- TypeScript
- Stelle
- 3.2k
- Fork
- 372
- Merge medio
- 1g 9h
- PR unite (30g)
- 229
Preparare l'ambiente
- Nessun Dockerfile né file Docker Compose
- Ha un modello di pull request
- Leggi la guida per i contributori
Come iniziare
- Leggi tutta la issue e poi la guida ai contributi del progetto.
- Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
- Fai un fork del repository e lavora su un branch.
- Apri una pull request che faccia riferimento al numero della issue.
Altre issue di TanStack/ai
-
openai-base: a Responses max_output_tokens cutoff ends as RUN_ERROR without usage instead of finishReason 'length'Forse già presa @pcbeingused333 l’ha presa 1 giorno fa. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 74/100
TanStack/ai#1916 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
[ai-harness] Rejected receipt resolves before durable rejection; recovery can rerun the inputAperta
Difficoltà 4/5 3-5 giorni Idoneità per principianti 52/100
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 3/5 1-2 giorni Idoneità per principianti 58/100
TanStack/ai#1924 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
ai-vercel-gateway: usage.cost leaves out gateway surcharges, and the BYOK credential type is not exposedForse già presa @pcbeingused333 l’ha presa 1 giorno fa. Aperta
Difficoltà 3/5 1-2 giorni Idoneità per principianti 55/100
I maintainer di solito rispondono entro 1 giorno
-
ai-gemini: native image generation throws "returned no id" when modelOptions.store is falseForse già presa @pcbeingused333 l’ha presa 1 giorno fa. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 57/100
I maintainer di solito rispondono entro 1 giorno
Issue simili
-
area:plugins area:security area:skills bug has repro
Difficoltà 2/5 1-3 ore Idoneità per principianti 76/100
anthropics/claude-code#105911 ·
I maintainer di solito rispondono entro 1 giorno
-
openspec-apply-change skill: when blocked, may select a ready artifact unrelated to missingArtifactsAperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
Fission-AI/OpenSpec#2165 ·
I maintainer di solito rispondono entro 1 giorno
-
🐞 bug 🧐 unconfirmed
Difficoltà 2/5 1-3 ore Idoneità per principianti 66/100
Tencent/tdesign-vue-next#7032 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 78/100
gnosis/gnosis_vpn#661 ·
I maintainer di solito rispondono entro 1 giorno
-
content-freshness: staleOwnedContent doc comments still say 'complete receipt' (follow-up to #7912)Apertatriage:bug-verified
Difficoltà 1/5 Meno di un'ora Idoneità per principianti 92/100
I maintainer di solito rispondono entro 1 giorno