This article is derived from the plugin's already-indexed fields (wiki / faq / compatibility_json / readme), not freshly generated by AI. Source field is noted at the end of each section.

Quick start

sandbase-harness

— source: plugin_wiki.wiki_content

Install & verify

dsh plugin --profile web add --allow-build=managed-agents github:sandbaseai/sandbase-harness

Run the command above in your DSH Web Profile. Then enable the plugin in the plugin list.

— source: plugins.install

Key points

• Claude Managed Agents-style /v1 API and local Console
• SQLite-backed agents, sessions, environments, credential vaults, memory
• local file/skill bytes stored in the workspace state directory
• Resumable Server-Sent Events for session replay and debugging
• One active model provider boundary configured through Settings V2

— source: plugin_wiki.readme_en (fallback readme_raw)

FAQ

Do I need to start any additional services after installing this plugin?

Yes. The plugin itself only registers the managed-agents-mcp bridge process in the DSH Web Profile. You must first start the underlying runtime using managed-agents start in another terminal (default listening on http://127.0.0.1:3000) for DSH to access the Agent and session via MCP tools.

What is the relationship between this plugin and DSH's official AI capabilities?

It is an independent local Agent runtime (based on SQLite + multi-sandbox backend), not a replacement for DSH's built-in models. DSH calls it as an external MCP service via the mcp__sandbase__* namespace, which then schedules OpenAI, Anthropic, or any OpenAI-compatible endpoints.

Where are sessions, artifacts, and credentials stored?

All stored in the .managed-agents/ directory under the workspace created when you run managed-agents init (SQLite file data.db, file bytes files/, skills packages skills/, sandbox snapshots snapshots/). The Bridge process does not persist any credentials.

Which model providers are supported?

Configure an active model provider boundary in Settings V2, covering OpenAI, Anthropic, and any OpenAI-compatible endpoints (README uses DeepSeek V4 as example). Specify the concrete model ID in the Agent YAML (e.g., gpt-4o, claude-sonnet-4-20250514, openai/gpt-5.5).

Is Docker required to use it?

No. By default, the Local sandbox executes commands using the current OS user and does not depend on Docker. You only need docker CLI or kubectl available when you switch the sandbox backend to docker or kubernetes in the Dashboard.

How to uninstall?

First stop DSH, then execute dsh plugin --profile web remove managed-agents, which will remove both profile dependencies and bundle injection layers. Runtime workspace data will not be automatically deleted, you need to manually clean up the .managed-agents/ directory.

What to do if "MCP startup failed" is reported at startup?

This indicates managed-agents-mcp is not in the PATH. Rebuild from source (npm ci && npm run build:runtime) and execute npm link, or check DSH startup logs to confirm if the mcp-sandbase-harness node is present.

How to troubleshoot 401 or 403 errors?

This is a runtime authentication mismatch. Check if the environment variable MANAGED_AGENTS_API_KEY matches the API Key configured on the Runtime side. Unauthenticated Runtimes are open by default (this variable is not needed).

— source: plugin_wiki.faq_json

Compatibility

• DSH: 未声明
• Node: >=22

— source: plugin_wiki.compatibility_json

Pitfalls

Review the upstream repo before installing. This guide is auto-derived from indexed fields and may lag the latest release. If anything contradicts the official docs, treat the upstream source as authoritative.

— source: general rule