Approve an agent action by replying to a text

Slack and Telegram approvals have buttons. SMS and iMessage do not, which is why approval-gated agents could not run there at all. They can now, by reading the decision out of the human's reply.

• The agent asks its question in plain language, and the platform always appends a mechanical action line naming the tool and its parameters, plus a contract line telling the person exactly how to answer. The tool name and parameters stay at least as prominent as the agent's own stated reason, which is context and never the deciding input.
• On iMessage, a tapback counts: a thumbs-up or thumbs-down on the approval message is correlated back to that specific prompt by message handle.
• Approvals are durable, so the run parks and resumes even across a redeploy.
• Binding an approval-requiring agent to a surface that cannot answer now raises a save-time warning instead of stranding runs at dispatch.