Today, a friend pinged me: one of his agents was stuck. He's not a SWE, so the error was opaque to him.

We share an isomux office, so I opened his agent's conversation and saw he tripped some Anthropic content filter. Likely a false positive.

Instead of context-switching into his problem myself, I went to one of my agents and typed:

/isomux-peer-review Sports analytics

(That's a skill that prompts an agent to review another agent's conversation and offer advice. "Sports analytics" is my friend's agent.)

My agent looked at the situation and produced this:

"How Nil can unblock:

Tell Pau to run /clear (most likely fix — the flagged content is now in context, so every turn re-trips the filter), then ask again with simpler phrasing."

My agent modeled (1) me, (2) my friend, (3) my friend's agent, and (4) the dynamic between all of us.

Shared-workspace agent collaboration is the future.