[bug] background subagents in auto mode flood me with permission prompts for safe…

Aperta
#2,701 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 5 giorni

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
18/100
Tipo di issue
Bug
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
bun, shell, typescript
Ambito
devtools, security

Direzione di ricerca

Start with the LifeOSHooks tool.call hook and the SubagentGitSandbox class with its wrapInline method, which the report says wrap every subagent Bash command in sandbox-exec. Check how the hook's changed input leads to the "gave no verdict" result in auto mode, and reproduce with a background subagent running ls. Done means safe read-only subagent commands are judged without reaching the human, while the git-internals protection still holds.

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

Descrizione

What happened

background subagents in auto mode flood me with permission prompts for safe commands, 15-20 clicks in a few minutes

Summary

In auto mode, a background subagent's Bash calls often get no verdict from the auto-mode reviewer and end up as permission prompts for the human, even for read-only commands.

Cause, from reading the code and the message Claude Code returns: the LifeOSHooks tool.call hook wraps every subagent Bash command in sandbox-exec (SubagentGitSandbox, wrapInline). Claude Code then reports "gave no verdict: a hook changed this call's input after the model wrote it, so the review was of different input from what would run", adds "This is not a judgment that the action is unsafe", and tells the model to issue the call again once.

Seen on 8.761.0 with four background subagents: about 15-20 prompts within a few minutes, for work that only read files. The same command type passes in the main session, which is not wrapped. It is intermittent (bun test passed while ls and bun Places.ts did not), and the agents copying the wrapped text into their retry adds more prompts.

Expected: the git-internals protection stays, and safe subagent commands are judged without reaching the human. The current workaround is telling subagents not to use Bash at all, which rules out any subagent that has to run tests, so unattended work is not practical.

Details

  • Kind: bug
  • LifeOS version: 8.761.0

Filed with the LifeOS /feedback command. The reporter approved this exact text before it was sent; paths, secrets and personal details were stripped locally.

Lingua principale
TypeScript
Stelle
19.9k
Fork
2.5k
Metriche di merge delle PR
Nessuna PR unita negli ultimi 30g

Preparare l'ambiente

Questo progetto non fornisce container di sviluppo, Dockerfile né guida per i contributori, quindi l'ambiente è a tuo carico: parti dal suo README e consulta la nostra guida al primo contributo per i passaggi generali.

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 danielmiessler/LifeOS

Tutte le issue di danielmiessler/LifeOS

Issue simili

Altre issue su TypeScript

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.