Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

(bug) create_app files do not appear in the agent's working folder on the local sandbox

Aperta
  • Qualcosa non va?
#7,422 1 commento 1 reazione 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
52/100
Tipo di issue
Bug
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
typescript
Ambito
backend

Direzione di ricerca

Start by tracing the create_app tool's write path to the session drive and how the local sandbox exposes that drive through its geesefs-mounted working folder. Reproduce the reported sequence and check whether files appear after a delay; the issue says Daytona and in-process sandboxes are not yet checked. Done means the starter files are visible in the agent's working folder when create_app returns, with the reproduction no longer requiring a hand-written copy.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

When an agent on a local sandbox builds an app the way the apps skill says (list apps/*/app.json, then call create_app(starter, "apps/<slug>")), create_app reports success but the files it wrote do not exist in the agent's working folder. The agent then rebuilds the app by hand, and says so to the user. I expect the starter files to be in the working folder as soon as create_app returns.

Seen on a v0.122.3 release stack (EE dev compose, one runner, local sandbox, Claude Code on Sonnet 5.5, store-backed working folder).

Steps

  1. Create an agent on the local sandbox, policy Allow all.
  2. Send: "Build me a small kanban board HTML app in your files using the board starter from your apps skill. Keep "access": "read-write" in its app.json so it can save. Tell me the path to its index.html when done."
  3. Watch the tool calls.

What happens

The session records, in order:

tool_call   Terminal  ls apps/*/app.json agent-files/.apps/registry/ agent-files/.apps/layout.json
tool_result           ls: cannot access 'apps/*/app.json': No such file or directory   (expected, nothing yet)
tool_call   mcp__agenta-tools__create_app  {"dir": "apps/kanban", "starter": "board"}
tool_result           {"paths": ["apps/kanban/app.json", "apps/kanban/index.html", "apps/kanban/config.json"], "template": "board@1"}
tool_call   Terminal  cd apps/kanban && cat app.json config.json; ls
tool_result           /bin/bash: line 1: cd: apps/kanban: No such file or directory
tool_call   Read File <cwd>/apps/kanban/app.json
tool_result           File does not exist.
tool_call   Terminal  sleep 5; ls -la apps
tool_result           ls: cannot access 'apps': No such file or directory
tool_call   Terminal  mkdir -p apps/kanban && cat > app.json ...   (the agent writes the board by hand)

The final reply tells the user: "create_app reported success, but its files did not show in my working folder. I wrote the board by hand with the same layout instead. It is a copy of the board starter, not the starter output."

The user ends up with a hand-written app instead of the maintained starter (board@1), and the agent spends extra turns and tokens (1.7M tokens for this one request).

Notes for debugging

  • create_app writes to the session's own drive (the cwd mount). After the turn, the drive API lists apps/kanban/{app.json,index.html,config.json,board.json}, so the objects exist in the store; the agent's later hand-written copy overwrote them, so the original starter output cannot be compared now.
  • The local sandbox's working folder is a geesefs mount of that drive prefix (--endpoint http://seaweedfs:8333 --no-detect --fsync-on-close, default cache settings).
  • The skill's first step (ls apps/*/app.json) runs before create_app, so the working folder has just looked up apps and found nothing.
  • Not checked yet: the Daytona and in-process sandboxes, and whether the files appear after a minute.
Lingua principale
TypeScript
Stelle
4.8k
Fork
690
Merge medio
20h 5m
PR unite (30g)
420

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di Agenta-AI/agenta

Tutte le issue di Agenta-AI/agenta

Issue simili

Altre issue su TypeScript

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.