This open-source project records model calls, retrievals, tools, and agent steps as spans that can flow to a team’s existing OTLP observability backend.
Why it is worth noticing
Teams already using Datadog, Grafana, or Jaeger can more easily see where an agent slows down, fails, or diverges from expectations.
What remains open
Real compatibility and correct minimization of sensitive prompts and outputs still need runtime testing.
It combines task ownership, terminal state, live code diffs, notifications, history, and tool permissions so several CLI agents can be supervised together.
Why it is worth noticing
Developers running several Codex, Claude Code, or other agents can poll fewer terminals and find tasks waiting for approval more easily.
What remains open
Turbo Mode is an auto-approval mechanism, not a security sandbox; whether it blocks high-risk actions still needs hands-on testing.
It offers visual editing, inline comments, and AI review while keeping saves, discussions, and publishing mapped to commits, pull requests, and merges.
Why it is worth noticing
PMs, designers, and engineers can author specs together without splitting documents and code across systems that drift apart.
What remains open
Its permissions, save behavior, and data flows in a real private repository have not yet been independently tested.
