oh-my-agent: managed skill caches and WikiSkill evolution

작성자

카테고리:

← 피드로
DEV Community · gracefullight · 2026-09-03 개발(SW)

gracefullight

The market, diagram, and video pipelines just moved to always-latest managed skill caches. When you orchestrate external tools, you want them at HEAD without dragging down your repository.

This week introduces a shared caching layer for third-party skills, meaning your agents always resolve the latest logic for generating architecture diagrams, market research, or video compositions.

What’s new

  • Managed skill layer: Introduced a shared caching system for external skills. Diagram, market, and video tools now resolve and update from GitHub Releases or upstream engines directly into local cache.
  • Diagram engine: Added oma diagram subcommands powered by the external archify skill for generating architecture diagrams.
  • Upstream market engine: Replaced the in-house market pipeline with the upstream last30days engine. The oma market run command now passes arguments straight to the source.
  • Dynamic video authoring: Stopped shipping vendored Remotion code. The new oma video compose command scaffolds compositions dynamically against the latest Remotion toolchain.
  • WikiSkill evolution: Added persistent rollout knowledge across optimization runs, complete with sandboxed live evaluations and validation gating.

Pushing skill resolution to runtime keeps the core CLI lean while giving agents the best available tools.

What’s fixed

  • Enforced literal UTF-8 output for Claude tool-call parameters to prevent unicode escape display corruption in interactive terminal prompts.
  • Disabled memory tools at the Serena MCP boundary to strictly enforce state safety during managed launches.

What’s better

  • State decoupling: Decoupled coordination state from Serena by routing session summaries through a neutral exporter.
  • Unified intelligence: Routed all managed clients through a single fixed Serena capability surface for context gathering.
  • Documentation structure: Moved the Quick Start section above the control boundary explanations in all README files to reduce time-to-action.

Installation

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/first-fluke/oh-my-agent/main/cli/install.sh | bash

Enter fullscreen mode Exit fullscreen mode

# Windows (PowerShell)
irm https://raw.githubusercontent.com/first-fluke/oh-my-agent/main/cli/install.ps1 | iex

Enter fullscreen mode Exit fullscreen mode

oh-my-agent is built for teams who orchestrate more than they prompt. Next up: exposing the managed skill layer to custom user workflows.

https://github.com/first-fluke/oh-my-agent

원문에서 계속 ↗