DEV Community

gracefullight
gracefullight

Posted on

oh-my-agent: session trajectories and skill compatibility checks

#ai #opensource #programming #productivity

Agent runs can span several CLIs, tool calls, and completion checks. Following that sequence means opening more than one log. This week's oh-my-agent changes add a session trajectory viewer, an HTML explanation renderer, and checks for how skills are read by different runtimes.

This is the delayed update for October 8, 2026. It covers changes available by 12:00 Australia/Sydney that day, through CLI 15.9.0. Later changes are outside this report.

What's new

  • Session trajectories. The viewer joins OMA lifecycle events with vendor transcripts. Terminal and dashboard views help you follow a session's sequence of messages, tool calls, and checks. The implementation was extended to eleven runtimes during the week. Implementation.
  • Markdown explanations. oma explain render turns a structured Markdown draft into a self-contained HTML page. Choose a panel grid or a document layout, then use oma explain lint to check the prose. This gives reviewers a file they can open without running the project. Implementation.
  • Skill compatibility checks. oma skills matrix plans checks for skill discovery, relative references, and final JSON output through native Claude and Codex CLIs. Installed-skill audits also support the body injection format used by agent-valley. These audits measure file access, not whether a skill can complete every task. Scope and commands.

The matrix defaults to planning. The following command prints a plan without starting vendor processes:

oma skills matrix --vendors codex --cases discovery,reference
Enter fullscreen mode Exit fullscreen mode

A live run requires --live and authenticated CLIs.

What's fixed

  • Schedule manifest updates are serialized to prevent concurrent writers from losing changes. Scheduled dispatch also resolves the CLI runtime.
  • Verification evidence is bound to the exact task and run receipts, so evidence from another run cannot stand in for the selected work.
  • Managed hook upgrades preserve user scripts, and uninstall preserves skills that lack OMA-managed ownership.

What's better

oma bridge loads only the bridge command instead of the full CLI command tree. Shared Serena daemon logs are bounded, and oma doctor warns when PATH contains different OMA versions. These changes address startup work, log growth, and version confusion.

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

If you use OMA across runtimes, start with a session trajectory to inspect what ran, then use the matrix to check how a selected runtime reads skill files.

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

Top comments (0)