Add OrcaReplay to Productivity Tools

[xizhuomengcontin]

Adds OrcaReplay to Productivity Tools, at the end of that section.

The section blurb is "tools, services which increase productivity, developer velocity and developer experience", and the neighbours are general developer CLIs — tenv, pyenv, kubefwd, Telert, claws. This is one of those: a CLI you put in front of an agent or a service so its runs stop being unrepeatable.

$ orca record claude -- claude -p "fix the failing test"
info recorded run=run_8f21c3 events=41 exit=0

$ orca replay last
info replaying exchanges=6 egress=blocked
info replay.done reused=6/6 exact=6 divergences=0 exit=0

The second command re-runs the session with no provider called and no API key needed — the model's answers come from the recording while your own code and tool calls run for real. Nothing is installed into the agent: it launches the process with the provider origin redirected for that process only.

Where the velocity comes from: a session that went wrong becomes something a colleague can reproduce without your credentials and without spending tokens, and the same recording is a zero-cost regression test in CI — which is otherwise impractical for anything with an LLM in the path.

Two limits worth stating rather than leaving in the footnotes:

• egress=blocked means model-provider egress, not network isolation. The recorded tool calls still execute for real, so it is not a sandbox.
• A matching replay shows the recorded run reproduces; it does not prove a fresh run would behave the same way, because the model is not re-asked.

Apache-2.0, Node 20+, npm i -g orcareplay. Disclosure: I maintain it.