When you tell an agent to build a complex feature, it usually needs a plan. When you do not provide one, the orchestrator should not just hand the request back to you.
Until this week, oma orchestrate required an existing plan file to proceed. Now, it generates one on the fly.
What's new
-
oma orchestratecreates a plan inline when no usable plan exists, using the same session ID so artifacts do not diverge. - Node 26 is now the required runtime for all workspaces and local shells.
- The OMA block enforces an absolute top-priority directive: execute imperative prompts verbatim and never build until explicitly asked.
- Qwen Code generation requests now run with a 5-minute timeout ceiling to prevent long agent turns from aborting mid-stream.
Global installations need to know where they live, not just where you are currently standing in the terminal.
What's fixed
-
oma doctor --globalandoma link --globalnow correctly resolve the global install root instead ofprocess.cwd(), fixing silent failures during global operations. - Serena search hits no longer truncate due to low character limits on initial installs.
- Removed short Korean keywords that falsely activated persistent modes and blocked session termination.
We also tweaked how the frontend skill handles default Next.js behaviors to protect your container CPU.
What's better
- The
oma-frontendskill defaultsnext/linkprefetch tofalse, cutting down on RSC payload requests for nav-heavy pages. - Over 40 vendor writers now use
safeGetInstallRoot(), making the CLI more resilient when.agentsdirectories are missing.
Installation
# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/first-fluke/oh-my-agent/main/cli/install.sh | bash
# Windows (PowerShell)
irm https://raw.githubusercontent.com/first-fluke/oh-my-agent/main/cli/install.ps1 | iex
oh-my-agent is built for teams who orchestrate more than they prompt. Next up: refined memory controls for long-running sessions.
Top comments (1)
Inline plan creation is interesting because it makes intent visible before execution. The next useful step is keeping the plan tied to evidence: what changed, what failed, and when the plan needs review.