Background sub-agents (general-purpose) get stuck indefinitely, never complete
- Lenguaje dominante
- Shell
- Estrellas
- 11.2k
- Forks
- 1.9k
- Merge medio
- 14 h 16 min
- PR fusionados (30 d)
- 6
Descripción
## Description
Background sub-agents launched via the `task` tool with `mode: "background"` and `agent_type: "general-purpose"` sometimes get stuck and never complete. The agent shows as "running" indefinitely. No completion notification arrives. `read_agent` with `wait: true` and a timeout returns the agent still running.
This doesn't happen every time, but it's frequent enough to be a real workflow blocker. The workaround is to abandon the stuck agent and re-launch as parallel `explore` agents instead, which reliably complete.
## Steps to Reproduce
1. Start an interactive Copilot CLI session
2. Launch a background general-purpose agent with a moderately complex prompt (e.g. scanning multiple directories, reading 20+ files, synthesising findings)
3. Wait for completion notification
## Expected Behaviour
The agent completes within a reasonable time (1-5 minutes) and sends a completion notification.
## Actual Behaviour
The agent stays in "running" state indefinitely. After 20+ minutes with no progress, the only option is to abandon it and work around the problem.
Notably, splitting the same work into 3 parallel `explore` agents completes in under 60 seconds each. The `general-purpose` agent type seems to hang on tasks that `explore` handles fine.
## Frequency
Observed in multiple sessions over the past week. Roughly 1 in 3 general-purpose background agents gets stuck. Explore agents have never gotten stuck for me.
## Environment
- Copilot CLI version: 1.0.34
- OS: macOS Tahoe 26.3.1 (Apple Silicon)
- Shell: zsh
- Terminal: Ghostty
## Workaround
Replace `general-purpose` background agents with parallel `explore` agents for read-heavy tasks. This reliably completes but limits you to Haiku and read-only tools.
Guía de contribución
Línea de trabajo
Start with the task tool's background mode for general-purpose agents and the read_agent wait path in an interactive Copilot CLI session. Reproduce the issue using a moderately complex, multi-directory prompt and compare it with parallel explore agents. Done means the background agent completes within a reasonable time and sends its completion notification instead of remaining running indefinitely.
Escrito por el modelo de indexación a partir del texto del issue.
Evaluación
- Stack tecnológico
- shell
- Área
- ai, cli
- Tipo de issue
- Error
- Dificultad
- 4/5
- Tiempo estimado
- 3-5 días
- Estado de actividad
- Tranquilo
- Claridad
- Necesita aclaración
- Aptitud para principiantes
- 35/100