Agenthood Release Digest: September 5, 2026
Date: September 5, 2026
Author: Agenthood Team
We've shipped a batch of releases across the Agenthood CLI, MCP integration, configuration, CI, plugins, and logging. Here's what's new.
Skills and CLI
Install skills from URLs and git repositories
- New command:
agenthood installfetches skills from URLs or git repositories. - Quality tiers: Skills now carry a quality tier — official, community, or experimental — with visual indicators in the CLI.
Publish command and dry-run installs
- Publish: Added
agenthood publishto publish skills. - Dry-run:
agenthood install --dry-runpreviews the installation without making changes. - Security fix: The installer now uses
execFileSyncinstead of a shell command to prevent shell injection.
Integrity state reporting
- Security:
agenthoodnow surfaces absent-lockfile and missing-skill integrity states, making it easier to diagnose incomplete skill installations.
MCP Integration
MCP server for skills
- New server: Agenthood now ships an MCP server that exposes skills as tools, allowing MCP clients to invoke skills directly.
- Refactor: Handlers were extracted for maintainability, the dependency was pinned, and documentation was updated.
Model Configuration
Default model updated
- Change: The default model is now
mimo-v2.5, replacingdeepseek-v4-flash. This affects new projects and configurations that don't specify a model.
CI and Code Review
Inline review findings
- CI: Analysis comments are now deduplicated, and review findings are posted inline on pull requests, reducing noise and putting feedback in context.
Plugin Ecosystem
Official opencode plugin
- Plugin: Added an official opencode plugin entry with a member-run tool.
Logging and Observability
Structured log API
- Logging: Added a structured log API and an
agenthood logcommand that reads from the trace store, giving you a unified view of agent activity.
What's Next
The recent releases point toward a continued focus on the skill lifecycle — installing, publishing, and verifying skills — as well as deeper MCP and editor integrations. We're also investing in observability and CI feedback loops. Expect more tooling around skill management and trace-based debugging in upcoming releases.