managed-agents

# SandBase Harness [English](./README.md) | [中文](./README.zh-CN.md) [](https://github.com/sandbaseai/sandbase-harness/stargazers) [](https://deepseek-plugin.org/plugins/sandbaseai/sandbase-harness) [](https://github.com/sandbaseai/sandbase-harness/releases/latest) [](https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.sandbaseai%2Fsandbase-harness) [](https://github.com/sandbaseai/sandbase-harness/discussions) [](https://github.com/sandbaseai/sandbase-harness/actions/workflows/codeql.yml) [](LICENSE) AI-readable project metadata: [llms.txt](./llms.txt) · [installation guide](./llms-install.md) A local-first runtime for AI agents. Sessions, sandboxed tools, memory, credentials, audit trails, and a built-in Console — all running on your machine or in your own infrastructure.  ## Why Agent SDKs handle the model loop. Production agents need more: persistent sessions, tool governance, sandbox boundaries, credential handling, memory, auditability, and a UI for humans to inspect what happened. `managed-agents` is that runtime layer — not a visual workflow builder and not another model SDK. | Need | What Harness provides | | --- | --- | | Run generated code safely | Local, Docker, Kubernetes, and self-hosted worker sandboxes | | Inspect long-running agents | Persistent sessions, resumable event streams, audit, and replay | | Control tool access | MCP toolsets, credential vaults, permission policies, and approvals | | Operate any model | OpenAI, Anthropic, MiniMax, and OpenAI-compatible providers, including DeepSeek V4 | | Keep infrastructure yours | Local-first SQLite and file storage with no required hosted control plane | ## Features - Claude Managed Agents-style `/v1` API and local Console - SQLite metadata by default for agents, sessions, environments, credential vaults, memory stores, files, skills, and API keys — local file/skill bytes in the workspace state directory - Resumable Server-Sent Events for session replay and debugging - One active model provider boundary configured through Settings V2 - Sandbox backends: local process, Docker (per-session containers), Kubernetes (kubectl exec/cp), self-hosted worker queue - MCP toolsets, permission policies, built-in tools, and skill packages - TypeScript SDK at `managed-agents/sdk` - Release gate: `npm run release:check` ## Quick Start Requirements: Node.js 22+, npm 10+, and a model provider API key (OpenAI, Anthropic, MiniMax, or any OpenAI-compatible endpoint). Docker is optional and only needed for Docker-backed sandboxes. ```bash git clone --branch v0.3.8 --depth 1 https://github.com/sandbaseai/sandbase-harness.git cd sandbase-harness npm ci npm run build mkdir ../my-agents && cd ../my-agents node ../sandbase-harness/dist/index.js init node ../sandbase-harness/dist/index.js start ``` `init` writes a workspace into the directory you run it from: an agent, a skills folder, and `config.yaml`, whose provider reference is the `${OPENAI_API_KEY}` environment variable. `start` serves the API and the Console on <http://127.0.0.1:3000>. Two steps finish the setup, both on **Settings > Setup** at <http://127.0.0.1:3000/dashboard>: 1. **The provider.** Paste your API key into the provider form and save. The page then reports that the saved configuration is not active yet, so restart the runtime — stop it with Ctrl+C and run the `start` command again, or use the restart button. A saved setting only takes effect at startup. If your provider is not in the list, choose the OpenAI-compatible vendor and set its base URL. 2. **The model.** In the **Agent models** panel, set the model ID your provider actually serves — `deepseek-chat` for DeepSeek, for example. An agent carries its own model ID, so the `gpt-4o` that `init` writes is not valid for every provider, and a wrong ID fails the turn with `model_not_found`. Send the first message from the Console: open **Sessions**, create a session for the agent, and type into the composer. From a terminal it is one command: ```bash node ../sandbase-harness/dist/index.js chat agent_assistant --message "hello" --tool-approval allow ``` `chat` sends that one message and exits once the turn settles; without `--message` it keeps the session open and streams until you interrupt it. `--tool-approval allow` preauthorizes the tool calls the agent may make, which the `init` template otherwise parks for approval and waits for a person to answer; see [CLI](#cli). The unscoped `managed-agents` name on npm is not this project. Until an official scoped package is announced in this repository, install only from the tagged GitHub source release shown above. Do not run `npx managed-agents` or `npm install managed-agents`. ### Try it in Codespaces [](https://codespaces.new/sandbaseai/sandbase-harness?quickstart=1) The included development container installs dependencies and builds the runtime. When the terminal is ready, start the server on the forwarded port: ```bash node dist/index.js start --host 0.0.0.0 ``` Open the forwarded **SandBase Harness Console** port, then configure a model in **Settings > Setup**. Codespaces usage may be billed by GitHub; the local quick start above remains free and keeps all runtime data on your machine. ## Screenshots | Console overview | Settings | API reference | | --- | --- | --- | |  |  |  | ## Use the Official SDK The runtime answers its own `/v1` API on that same port, and an official Anthropic TypeScript SDK client drives it unchanged: point the client's `baseURL` at the runtime, give it the runtime API key, and the quickstart in [`examples/official-sdk`](examples/official-sdk/README.md) runs a whole turn — message, tool call, tool result, final reply — against it. That example is executed on every pull request by `tests/conformance/official-sdk-quickstart.test.ts`, so the compatibility it describes is compatibility that is tested rather than claimed. The same surface is specified in [docs/api.md](docs/api.md), and this repository's own TypeScript SDK is documented under [SDK](#sdk) below. ## CMA compatibility Coverage of the published Claude Managed Agents contract is declared entry by entry in [`src/core/capabilities/matrix.ts`](src/core/capabilities/matrix.ts): of the official SDK's route surface (110 method/path pairs as of SDK `0.131.0`), 90 routes are mounted and 20 refuse by name; no route is deferred. `Partial` and `Unsupported` entries always name their reason. The table below is generated by `npm run docs:compat`, and a contract-honesty test fails when it drifts from the matrix. <details> <summary><strong>Full compatibility table (generated)</strong></summary> <!-- compat-table:start --> | Area | Official capability | Status | Notes | | --- | --- | --- | --- | | headers | `compatibility-header-admission` | Partial | Version, beta, and mutual-exclusion rules are enforced for any request that carries a compatibility header. A request with no compatibility header is accepted as a local caller, which the published contract does not define; this header-free path is a deliberate local-first extension for a self-hosted single-tenant runtime, recorded as such in the headers contract §4, and it is why this entry stays partial. | | headers | `extension-namespace-exclusion` | Supported | | | pagination | `opaque-cursors` | Partial | A collection's envelope follows the mount: the operations router serves `/v1` with `{data, prev_page, next_page}` and its `/v1/x` mirror with the local `{data, has_more, first_id, last_id}`, chosen through one pager so no handler emits both spellings, and cursors that are readable base64url JSON rather than opaque binary. Every canonical `/v1` collection serves the canonical envelope, with no exceptions: the complete-set listings carry `{data, prev_page: null, next_page: null}` and the windowed ones carry a followable cursor — `/v1/sessions` pages by number under `{order, created_at bounds, page}` and rejects a cursor replayed under another ordering or creation window, `/v1/sessions/{id}/events` carries `{order, filter, after_id}`, `/v1/skills` and the credential audit listings use `{offset, filter}` — so a cut page says so instead of looking complete. The one listing that is neither shape is `/v1/environments/{id}/work-items`, a windowed extension that adds a `counts` object and is named in the contract. | | pagination | `cursor-query-binding` | Supported | | | errors | `structured-error-envelope` | Supported | | | agents | `agent-crud` | Supported | | | agents | `model-object-profile` | Partial | String and object model forms parse field by field. `effort` and `speed` are stored, returned by the read projection (the agent read, the version read, and the session snapshot), and executed on the Anthropic provider under a model capability table — `effort` becomes `output_config.effort`, `fast` becomes `speed: "fast"` with the fast-mode beta, and adaptive-thinking models receive `thinking: {type: "adaptive", display: "omitted"}`; a listed model refused a level or speed it cannot take fails admission, an unknown model id or non-Anthropic provider sends nothing, and a deployment's own `reasoning_effort` model setting is operator-level and separate. `inference_geo` is refused by name with `unsupported_model_field` because this runtime has no inference-geography control; and a canonical `multiagent` roster is refused by name rather than executed. | | agents | `multiagent-roster` | Unsupported | A canonical `multiagent` roster is refused by name on both agent create and agent update, because no thread, coordinator, or advisor surface exists to honour it; accepting it would let a caller believe delegation by roster is in effect. Local delegation is registered separately as an extension. | | agents | `local-delegation-subagent` | Supported | | | sessions | `session-lifecycle` | Supported | | | sessions | `initial-events` | Supported | | | sessions | `prompt-caching` | Supported | | | sessions | `session-update` | Supported | | | budget | `session-budget` | Partial | Consumption is priced in integer microcents from the append-only log, and a session may declare a max_list_cost ceiling at creation. The builtin loop checks the ceiling inside a turn: the step that crossed the cap is the last one, the session idles with stop_reason budget_reached and a session.usage immediately before it, and an accepted budget update or removal resumes the session on its own — a tool call the ceiling stranded is settled so the resumed turn sees a paired transcript. At the ceiling the next work-starting event is refused with budget_reached while events that settle work already in flight are still accepted, so the next model request does not start; a declared outcome's revision loop reads the same spend and stops at the ceiling too, closing the outcome with result budget_reached rather than starting another grading pass or turn, because the loop's turns are internal to an event that was already admitted. Two deviations are deliberate: prices come from an operator-supplied cost profile rather than official list prices, so a session whose model the profile cannot price is refused a budget and usage.list_cost is withheld while any used model is unpriced; and the pause is reported on the session's own status_idle only, because the published thread-level budget_reached signal belongs to the thread surface, which this runtime does not implement. | | events | `append-only-event-log` | Supported | | | events | `processed-at-lifecycle` | Supported | | | events | `session-error-structure` | Supported | | | events | `error-enum-completeness` | Supported | | | events | `model-request-span-pair` | Supported | | | streaming | `resumable-sse` | Supported | | | streaming | `agent-message-stream-preview` | Supported | | | tools | `builtin-tool-execution` | Supported | | | tools | `web-fetch-execution` | Partial | WebFetch executes over HTTP/HTTPS with domain policy, per-redirect revalidation, private-address rejection, timeout and byte caps, HTML text extraction, and a max_content_tokens budget; it converts text-like content only (no image or PDF rendering), the token budget is a character estimate, and TLS hostnames are verified but content is not sandboxed beyond redaction. | | tools | `web-tool-domain-policy` | Supported | | | tools | `tool-output-overflow` | Supported | | | tools | `mcp-tool-approval-gate` | Supported | | | tools | `auto-permission-policy` | Supported | | | custom-tools | `custom-tool-declaration` | Supported | | | custom-tools | `custom-tool-worker-execution` | Supported | | | system-message | `system-message-events` | Supported | | | memory-stores | `memory-crud` | Supported | | | memory-stores | `memory-limits-and-preconditions` | Supported | | | memory-stores | `memory-version-audit` | Supported | | | memory-stores | `memory-multi-mount` | Supported | | | memory-stores | `memory-worker-materialization` | Supported | | | github-repository | `github-repository-materialization` | Supported | | | github-repository | `github-repository-identity-freeze` | Supported | | | files | `file-resources` | Supported | | | files | `file-mount-path` | Supported | | | credentials | `canonical-credential-wire-profile` | Supported | | | credentials | `credential-rotation` | Supported | | | credentials | `credential-injection-execution` | Supported | | | credentials | `oauth-refresh` | Supported | | | credentials | `mcp-oauth-validation` | Supported | | | operations | `webhook-subscriptions` | Partial | Locally implemented, but the delivery behaviour is not the published contract. Subscriptions are managed over REST under /v1/webhooks (with the /v1/x mirror) and delivery runs from a bridge the runtime composes at startup: each durable event is projected as it is broadcast and a 60-second tick retries due deliveries and runs due deployments, while POST /v1/webhooks/dispatch and POST /v1/webhooks/retry-due remain for on-demand passes. Every attempt carries the published header names and a Standard Webhooks v1 signature over id.timestamp.body — a retry keeps the event id and signs with its own timestamp, and each subscription holds its own whsec_ secret that is returned once at creation, and a rotation window keeps the previous secret valid in a second webhook-signature entry until it is retired — manually by retire-secret, or automatically once the window has been open for the duration the deployment set (24 hours by default, settable with MANAGED_AGENTS_WEBHOOK_ROTATION_WINDOW_SECONDS and recorded at startup as webhook_rotation_window; expiry drops the previous columns, enforced where a signature is produced and swept on the retry tick, while a window opened before the since timestamp existed keeps manual-retire behaviour) — the payload is the published {type: "event", id, created_at, data: {type, id, organization_id, workspace_id}} reference envelope with webhook-id equal to the event id and local constant org/workspace values, subscriptions may only name events from the official catalog — *, prefix.*, and unknown names are refused at write time — and the session stream reaches subscribers only through the published-name projection (status events mapped, budget_reached deduplicated per session and ceiling, internal events dropped) while resource routes publish the lifecycle events for sessions, agents, environments, vaults and credentials, memory stores, and deployments; every catalog name has a producer (the coarse session lifecycle names ride the same transitions: pending at creation, running/idled/requires_action through the stream projection), the published names with no producing surface (session.thread_*, agent.deleted) are refused at subscription like any unknown name, of the three published auto-disable cases all three exist, two unconditionally and one opt-in (an attempt that observes a redirect disables the endpoint with the published disabled_reason and is never retried; an attempt whose host is an internal name or resolves to a private address is refused before any connection with its own published reason but only when the deployment sets MANAGED_AGENTS_WEBHOOK_SCREEN_PRIVATE_ADDRESSES, off by default because loopback is private and a self-hosted receiver normally shares the host; and an endpoint failing without interruption for at least a window is disabled with the published sustained-failure reason, where the contract publishes the trigger shape — duration, not attempt count, with a 2xx resetting it — but no length, so the window is a local parameter: it defaults to 10 minutes, a deployment sets its own with MANAGED_AGENTS_WEBHOOK_SUSTAINED_FAILURE_WINDOW_SECONDS, and the runtime records the window in force at startup because a deployment variable has no write path of its own). Retries follow the published jittered 5-120s exponential backoff: the ceiling doubles from 60s to 120s and each delay is drawn uniformly between 5s and that ceiling. | | operations | `scheduled-deployment-timers` | Partial | The published deployment surface end to end: the object answers with type deployment, a depl_ id, a pinned {type, id, version} agent, environment_id, a required non-empty initial_events list (the session admission plus the deployment-only system.message), resources, vault_ids, budget, metadata, and a schedule object with last_run_at and the next three upcoming_runs_at — null for a manual-only deployment, because cron is nullable since M052. Both mount spellings serve create/read/update (POST the published verb, PUT the local one)/archive/pause/unpause/run/run-due from one router, and the local flat aliases (agent_id, cron, timezone, payload) remain accepted. Each run creates its session through createWithInitialEvents and records a drun_ run readable at /v1/deployment_runs (deployment_id, has_error, trigger_type, created_at filters) with trigger_context carrying scheduled_at for a timed run. Failure is asymmetric per the published contract: a missing or archived bound agent archives the deployment with no run, a recoverable session_rate_limited_error records a failed run only, and other classified failures record the run and auto-pause the deployment with paused_reason.error mirroring the run's classified error.type. Manual pause and unpause exist and a paused deployment still accepts a manual run; timed runs publish deployment_run.started/.succeeded/.failed and manual runs publish none, while deployment.created/.updated/.paused/.unpaused/.archived publish on their transitions including the agent-gone cascade, and DELETE removes the deployment and its run records in one transaction then publishes deployment.deleted. The runtime's 60-second tick runs due deployments and startup re-arms their forward schedule without replaying a missed trigger. Remaining gap: mcp_egress_blocked_error has no producing path because MCP egress is not gated. | | operations | `outcome-grading` | Supported | | | operations | `outcome-evaluation` | Supported | | | capabilities | `capability-inventory-endpoint` | Supported | | | capabilities | `capability-status-truthfulness` | Supported | | | environments | `environment-hosting-config` | Supported | | | environments | `environment-network-policy` | Partial | The published config.networking object is accepted in its own vocabulary (limited/unrestricted, allowed_hosts, allow_mcp_servers, allow_package_managers) and normalized into the recorded local config.network spelling by one normalizer, with fail-closed defaults for an unrecognized type and for an unset permission, and a request declaring both spellings inconsistently refused. A limited policy is now applied: every limited sandbox gets a per-session loopback egress proxy that speaks CONNECT and absolute-URI HTTP and admits only the effective allowlist (allowed_hosts plus, when the package-manager flag is set, the curated public registry endpoints). The docker provider attaches the session container to an --internal network whose only permitted peer is a relay sidecar forwarding to the proxy, so the boundary is enforced rather than advisory; the local provider injects the proxy variables into every sandbox subprocess and stdio MCP server, which is advisory by construction — a process that ignores proxy variables egresses freely — and reports best_effort rather than claiming enforcement. The MCP connect boundary refuses a url server whose host the policy does not cover (unless allow_mcp_servers is set), web_fetch intersects the declared allowlist with its existing domain and SSRF guards at every redirect hop, and the Environment read projects networking_enforcement (enforced / best_effort / unsupported / not_applicable) from the effective backend's declared capability. It remains partial because the kubernetes and self-hosted providers install no egress boundary — a limited policy on them is reported as unsupported rather than applied — and because the local provider's enforcement is advisory by nature. | | environments | `environment-work` | Partial | The entire published Work API is mounted over the local tool-execution queue. The data plane: poll claims the oldest claimable item scoped to the calling credential's environment and returns it in the published BetaSelfHostedWork shape with a per-claim secret (base64url JSON carrying a sessions_token minted for that claim, plus api_base_url); ack commits the claim, heartbeat renews the heartbeat lease with the published NO_HEARTBEAT first-claim sentinel and expected_last_heartbeat optimistic-concurrency check (412 carrying the server's current_state), update merges a metadata patch, and stop records the queue's stop marker. The management plane: list pages items newest-first under a keyset page cursor, retrieve answers one item through the same item-scope fence as the item verbs, and stats reports the published work_queue_stats fields computed from the lease model (depth = claimable now, pending = claimed inside its lease, workers_polling = identities seen on poll in 30s). The sessions_token is also the worker's session-level credential beyond the work family: it retrieves its own session (the resources list is how the worker discovers attached stores), lists and streams its events, posts only the tool-answer event types, and reads or writes the memories of stores the session attached — with access: "read_only" attachments refusing writes with 403 and every other route answering 401. The projection is honest about its edges: data is always {type: "session", id} because every local item belongs to a session, per-item desired_ttl_seconds is reported back rather than applied, force-stop has no distinct local mode, and result reporting stays on the local /v1/x/worker channel. Remains partial for those semantic deltas — not for missing routes; nothing in the family is a refusal. | | routes | `documented-route-surface` | Supported | | | dreams | `dreams` | Supported | | | threads | `threads-and-coordinator` | Unsupported | Not implemented: there is no thread resource, no thread lifecycle or per-thread event isolation, no coordinator or advisor role, and no thread-scoped budget event. The five published thread routes are mounted `unsupported_capability` refusals naming this capability, so an SDK caller decodes a 400 rather than hitting a 404. A request carrying a `multiagent` roster is refused by name rather than silently stripped. Delegation exists only as the local single-level `delegations` / `enable_general_subagent` extension, which is not this surface. | | unsupported | `session-budget-alerts` | Unsupported | (not_applicable) Budget notification is a hosted billing feature: it needs an outbound channel to a party who pays for the account, and SandBase is single-tenant and local, so the operator is already the only party to notify. | | unsupported | `mcp-tunnel` | Unsupported | (not_applicable) MCP tunnel is a hosted connectivity feature outside the local-first scope. | | tools | `web-search-execution` | Supported | | <!-- compat-table:end --> </details> ## CLI ```bash managed-agents init managed-agents start [--host 127.0.0.1] [--port 3000] managed-agents apply <file...> # create or update agents/environments/deployments from manifests managed-agents list managed-agents reload managed-agents chat <agent-id> --message "hello" [--tool-approval ask|allow|deny] managed-agents template list | install <name> | create <name> ``` `apply` is the declarative form of the same API: a Markdown file whose frontmatter is the agent definition (the body below `---` becomes the system prompt), a `.yaml` environment document, or a Markdown deployment manifest (frontmatter carries `name`, `agent`, `environment`, `schedule`; the body becomes the initial `user.message`). The resolved ids land in `managed-agents-lock.json` beside the files, so re-running `apply` after an edit updates the same resource instead of creating a duplicate — commit the lock file with the manifests. A later file may reference an earlier one by name, so `apply coding-assistant.md environment.yaml deployment.md` wires the deployment to the rows the first two files created. A turn whose tool needs approval parks instead of failing, and `chat` asks before running it, then lets the runtime continue the same turn. `--tool-approval allow` decides every such call in advance, which is what a script or a CI job uses, and `deny` refuses them. With no terminal to prompt, the default `ask` answers nothing and exits non-zero with the calls that are waiting named, so a script states its policy rather than inheriting one. A custom tool is the exception: only your own client can produce its result, and `chat` says so and exits non-zero. See [usage](docs/usage.md#cli-commands). ## SDK ```typescript import { ManagedAgentsClient } from 'managed-agents/sdk'; const client = new ManagedAgentsClient({ baseUrl: 'http://127.0.0.1:3000', }); const session = await client.sessions.create({ agent: 'agent_...', environment_id: 'env_...', }); for await (const event of client.sessions.chat(session.id, 'Hello')) { if (event.type === 'agent.message_chunk') { process.stdout.write(event.delta ?? ''); } } ``` The `/v1` API follows Claude Managed Agents resource shapes, so you can also point the Anthropic SDK at the local runtime: ```typescript import Anthropic from '@anthropic-ai/sdk'; const client = new Anthropic({ apiKey: process.env.MANAGED_AGENTS_API_KEY ?? 'local-dev-key', baseURL: 'http://127.0.0.1:3000', }); const session = await client.beta.sessions.create({ agent: 'agent_...', environment_id: 'env_...', }); ``` ## Authentication Open by default. Authentication activates when at least one API key exists: ```bash # Static key via environment export MANAGED_AGENTS_API_KEY=sk-local-example # Or create a managed key curl -X POST http://127.0.0.1:3000/v1/api-keys \ -H "Content-Type: application/json" \ -d '{ "name": "Local Console" }' ``` Clients send `Authorization: Bearer <key>`. ## Integrations and examples - **DeepSeek Harness plugin** — run this runtime as a DSH plugin over MCP stdio: install, preflight, tool list, and troubleshooting live in [`examples/deepseek-harness`](examples/deepseek-harness/README.md). A DSH project can also take a portable Skill from GitHub source — `npx --yes github:sandbaseai/sandbase-skills add multi-source-search` installs into `.dsh/skills/multi-source-search`. - **Agent Plugins 1.0 clients** (Copilot CLI, VS Code) and the standalone **MCP bridge container** — see [`agent-plugin/PLUGIN.md`](agent-plugin/PLUGIN.md). - **Use cases** — the [Showcase](docs/showcase.md) walks through an auditable coding agent, DSH as an interactive front end, and controlled code execution across Local, Docker, Kubernetes, and self-hosted sandboxes. - **Agent configuration** — the YAML agent definition, `config.yaml`, and the workspace layout live in the [usage guide](docs/usage.md); curl walkthroughs for every resource are in [docs/api.md](docs/api.md). ## Documentation - [Machine-readable project metadata](llms.txt) - [Agent / MCP installation guide](llms-install.md) - [Agent Plugin marketplace manifest](agent-plugin/PLUGIN.md) - [Installation](docs/installation.md) - [Usage Guide](docs/usage.md) - [API Reference](docs/api.md) - [Skills](docs/skills.md) - [DeepSeek V4](docs/deepseek-v4.md) - [MiniMax](docs/minimax.md) - [Deployment](docs/deployment.md) - [Architecture](docs/spec/architecture.md) - [DeepSeek Harness integration](examples/deepseek-harness/README.md) - [Contributing](CONTRIBUTING.md) - [Citation metadata](CITATION.cff) - [Changelog](CHANGELOG.md) ## Development ```bash npm ci npm run typecheck # src + tests + Console npm test # vitest npm run build # runtime + console + SDK npm run release:check # full local release gate ``` `release:check` runs typecheck, tests, both builds, `npm pack --dry-run`, CLI init smoke, and `examples/basic` startup smoke. ## Star and share If this runtime solves a real agent-infrastructure problem for you, [star the repository](https://github.com/sandbaseai/sandbase-harness) so other builders can find it. Ecosystem directories, community guides, and related projects are in [docs/ecosystem.md](docs/ecosystem.md). Community use-case discussions: [memory migration between Codex, Claude Code, and DSH](https://github.com/deepseek-ai/deepseek-harness/discussions/14#discussioncomment-18202967), [sandbox and filesystem protection for third-party plugins](https://github.com/deepseek-ai/deepseek-harness/discussions/5068#discussioncomment-18202943). ## License [Apache-2.0](LICENSE)