AI news · October 8, 2026
Docker Agent v1.149.0 adds GitHub skills and a Decisions evaluator
Docker Agent v1.149.0 turns agent setup into YAML, supports multiple model providers and MCP servers, and now loads skills from GitHub repositories. Docker Desktop 4.63+ includes it.
- latest release on Oct 7
- v1.149.0
- Docker Desktop version
- 4.63+
- listed model-provider families
- 5+
Docker Agent is an AI agent builder and runtime that uses declarative YAML instead of application code for the first setup. The project supports multi-agent flows, MCP servers, several model providers, retrieval options, and sharing through OCI artifacts. The v1.149.0 release on October 7 added GitHub repository skill loading, evaluator routing through the models gateway, and an OpenAI Decisions evaluator backend, alongside debugging and pricing fixes. Docker Desktop 4.63 and later includes the agent tooling.
The useful part is the operational surface: a team can describe an agent, its tools, evaluator, and retrieval settings in a file that can be reviewed and shared. That does not make the workflow safe by itself. Builders still need limits on tool access, clear test data, and a way to inspect failed runs. Start with one read-only task and keep the agent's permissions smaller than the task's possible impact.
What you can do with it
Try Docker Agent from the public repository or Docker Desktop 4.63+. Define one read-only agent in YAML, add one MCP server, and run a fixed test set. Inspect tool calls, evaluator results, and failure recovery before adding write access or multiple agents.
Our take
Docker is moving agent work toward a familiar build-and-package workflow. The GitHub skill feature and Decisions evaluator are more useful than another demo because they affect repeatability. The product still needs strong defaults for secrets, tool permissions, and evaluation before it can be a safe team standard.
Links Try today Release notes HN discussion
Source: Docker ↗ — Made With Models writes the brief; the reporting is theirs.