mirror of
https://github.com/openclaw/openclaw.git
synced 2026-08-25 03:45:46 -06:00
Vincent Koc
27 lines
1.2 KiB
YAML
27 lines
1.2 KiB
YAML
title: Agent sandboxed exec behavior
|
|
|
|
scenario:
|
|
id: agent-sandboxed-exec-behavior
|
|
surface: runtime-tools
|
|
coverage:
|
|
primary:
|
|
- agent-runtime.sandboxed-exec-behavior
|
|
objective: Verify host:auto executes through the real Docker sandbox selected for an agent attempt.
|
|
successCriteria:
|
|
- The attempt-level resolver provisions a real session-scoped Docker sandbox.
|
|
- The public coding-tool surface routes exec host:auto into that sandbox.
|
|
- The command observes Docker and the /workspace mount and persists a workspace marker.
|
|
- A host executable outside the workspace remains inaccessible and is never executed.
|
|
- The sandbox runtime and isolated state are removed after the proof.
|
|
docsRefs:
|
|
- docs/gateway/sandboxing.md
|
|
codeRefs:
|
|
- src/agents/embedded-agent-runner/run/attempt-setup.ts
|
|
- src/agents/agent-tools.ts
|
|
- src/agents/bash-tools.exec-run.ts
|
|
- test/e2e/qa-lab/runtime/agent-sandboxed-exec-behavior.e2e.test.ts
|
|
execution:
|
|
kind: vitest
|
|
path: test/e2e/qa-lab/runtime/agent-sandboxed-exec-behavior.e2e.test.ts
|
|
summary: Provision Docker, construct agent tools, and execute a host:auto command across the sandbox boundary.
|