Gemini Enterprise Managed Agents
Manage stateful Gemini Enterprise agents through the Managed Agents API.
The document identifies ADC Bearer-token authentication and calls out update_mask and MCP header scope. However, examples permit unrestricted networking ("*"), include filesystem and external MCP tools, and cover resource creation, mutation, and deletion. There is no least-privilege IAM guidance, per-operation user confirmation, sensitive-data handling, delete confirmation, rollback, or recovery procedure, so substantial points are deducted.
The CRUD, long-running-operation polling, and data-plane handoff are broadly coherent and the happy path is understandable. Error responses, retries, timeouts, pagination, concurrent updates, permission failures, and abnormal-input handling are absent, and the examples' API version, fields, and model availability are not verified. The static cap keeps this at 7 rather than above 10.
The audience and primary scenario—managing stateful Agents through REST—are reasonably clear, and the document points interaction work to a separate skill. It does not define non-fit cases, required IAM/API enablement, regional constraints, or precise trigger boundaries. Core functionality depends on Google Cloud services without any mainland-China reachability or alternative-path guidance, so points are deducted.
The structure is readable and includes installation entry points, setup, examples, licensing, and repository support/contribution paths. The skill lacks its own version, changelog, named maintenance responsibility, dependency checklist, FAQ, limitation disclosure, and troubleshooting guidance; Apache-2.0 is established only at repository level, so full marks are not justified.
It supplies directly usable curl examples for create, poll, get, list, update, and delete, plus Python/JSON snippets linking to the data plane. These cover the core task, but the examples are templates without verified end-to-end outputs, failure handling, or complete prerequisite and permission guidance. Considerable user validation and supplementation remain necessary, so the score stays within the static cap.
The evidence is limited to the target SKILL.md, repository README, and LICENSE. It provides concrete endpoints and request examples but no committed tests, CI coverage, official-document citations, or third-party execution evidence. Key paths therefore cannot be independently reproduced from the supplied files, warranting only a low score.
- Do not reuse the "*" network allowlist or place real secrets in examples; restrict domains, review MCP services, and inject secrets through a managed secret store.
- Updates and deletion change external state; confirm the target, back up configuration, and establish rollback or recovery procedures before execution.
- Before production use, verify the v1beta1 endpoint, fields, regional support, IAM permissions, and model identifier; the document provides neither error handling nor executable tests.
- Mainland-China users may not have reliable access to the required Google Cloud services; validate reachability and compliance first.
What does this skill do, and when should you use it?
This skill guides developers through the Gemini Enterprise Agent Platform Managed Agents API for custom, stateful Agent resources. It covers creation, retrieval, listing, patching, deletion, and polling long-running operations. Agents can be configured with system instructions, tools, sandboxed files, skill registries, remote environments, and mounted Cloud Storage resources. Conversation execution belongs to the Data Plane and should use the separate gemini-interactions-api skill.
Sets project, location, and ADC access-token variables; calls REST Control Plane endpoints to create, get, list, update, and delete Agents; polls creation operations; configures code execution, filesystem, Google Search, URL Context, and MCP tools; mounts Cloud Storage or Skill Registry resources into an Agent workspace; and uses update_mask for targeted configuration changes.
- A Google Cloud developer needs to provision a custom Agent with system instructions and tools.
- A platform engineer needs to retrieve or list configured Agents in a project.
- A team needs to patch an Agent's instructions, tools, or mounted resources in place.
- A developer needs to mount files, folders, or skills from Cloud Storage into an Agent container.
- An integration team needs to configure a third-party MCP server for routed tool requests.
What are this skill's strengths and limitations?
- Covers the full Managed Agents Control Plane CRUD workflow.
- Provides copyable curl requests, endpoints, and JSON payload examples.
- Supports Cloud Storage, Skill Registry, tools, and third-party MCP configuration.
- Documents long-running creation operations, MCP routing, and the update_mask requirement.
- The instructions target Gemini Enterprise Agent Platform and Google Cloud; applicability to other Agent platforms is not evidenced.
- Requires a Google Cloud project, ADC credentials, and the gcloud CLI; required IAM roles are not documented.
- No test suite, version-compatibility matrix, pricing information, or failure-retry strategy is provided.
- Conversation and streaming workflows are outside this skill's scope and require gemini-interactions-api.
How do you install this skill?
Install the collection with: npx skills add google/skills. The installer allows selection of individual skills from the repository; the source does not document the local destination or client-specific loading steps after selection.
How do you use this skill?
In an Agent Skills-compatible client, request a management task such as “Create an Agent named my-custom-agent with a system instruction and a mounted Cloud Storage skill directory.” Actual REST calls require PROJECT_ID, LOCATION, and ACCESS_TOKEN=$(gcloud auth print-access-token), using the Agents Control Plane endpoints shown in the skill. For conversations or streaming, consult gemini-interactions-api.
How does this skill compare with similar options?
This skill manages Agent resources in the Control Plane; the gemini-interactions-api skill handles conversations, tool execution, and streaming with provisioned Agents.