Environment schemas

​

Environment object

​

Environment config

{"type": "self_hosted", "setup_script": "./initialize-worker.sh"}

​

Environment packages

​

Environment setup script

{
  "config": {
    "type": "cloud",
    "setup_script": "set -euo pipefail\n[ -d /data/workspace/repo/.git ] || git clone https://github.com/me/repo /data/workspace/repo\ncd /data/workspace/repo && pnpm install --frozen-lockfile"
  }
}

​

Environment startup script

• The script must exit. Run long-lived processes in the background with their output redirected to a file, for example nohup pnpm dev > /tmp/dev.log 2>&1 &; they keep running after the script exits. A command that runs in the foreground and never exits blocks tool calls until the timeout.
• The script exiting does not mean the service is ready: the agent's first tool call starts as soon as the script exits, while a service started in the background may take a few seconds to listen. When the agent needs a service right away, wait for its port at the end of the script before exiting, for example for i in $(seq 30); do curl -sf -o /dev/null localhost:3000 && break; sleep 1; done.
• At the timeout, the script is stopped together with the background processes it started, and the run counts as failed.
• On a resume, processes from before may still be running (some runtimes keep memory across a pause), so the script must handle a service that is already running, for example pgrep -x redis-server || redis-server --daemonize yes. Do not check with pgrep -f for text that also appears in the script: the shell running the script carries the script's text in its command line and matches, so the service is never started; match the process name with pgrep -x, or check the service's port.
• It runs at most once per start: the run is recorded before the script starts, so a run cut short is not repeated in the same start; it runs again on the next start.

{
  "config": {
    "type": "cloud",
    "packages": {"apt": ["redis-server"]},
    "startup_script": "pgrep -x redis-server || redis-server --daemonize yes\ncurl -s -o /dev/null localhost:3000 || (cd /data/workspace/repo && nohup pnpm dev > /tmp/dev.log 2>&1 &)"
  }
}

​

Environment storage

{
  "type": "cloud",
  "storage": {
    "buckets": [
      {
        "name": "datasets",
        "provider": "aliyun-oss",
        "region": "cn-hangzhou",
        "bucket": "my-bucket",
        "prefix": "team/data/",
        "access": "read_only",
        "credential": {"type": "ram_role", "role_arn": "acs:ram::1234567890123456:role/qoder-oss-mount"}
      }
    ]
  }
}