Publicado en August 20, 2026

A pause between parts 7 and 8: four places where running nexo_mail against a real inbox forced the design past what the earlier posts described.

Publicado en August 11, 2026

Why the merge, dedupe and narration of three email extractions lives in a Nexo skill rather than in Ruby, and the seam for fixing model arithmetic.

Publicado en August 6, 2026

Nexo's four sandbox tiers and four permission modes, and how an email agent gets write access to one folder and nothing else, enforced in code.

Publicado en July 30, 2026

Fan three email source agents out with Nexo.concurrent: bounded concurrency, ordered results, and a sequential fallback when the async gem is missing.

Publicado en July 29, 2026

Wrap three email source agents in a Nexo workflow that skips what it cannot run, records every step in an event log, and still finishes when a source fails.

Publicado en July 28, 2026

Read Gmail over IMAP and HEY through its CLI with custom RubyLLM tools, attached to a Nexo agent that stays read-only by construction, not by prompt.

Publicado en July 27, 2026

Move an email agent's triage policy from a Ruby heredoc into a Markdown skill, and why a skill adds instructions only and never widens what the agent can do.

Publicado en July 25, 2026

bin/ci green has always been necessary and not sufficient. With AI coding, one session writes both the implementation and the test that checks it.

Publicado en July 23, 2026

Connect a Nexo agent to Apple Mail over MCP and triage an inbox read-only by construction, where a fail-closed allow-list makes send and delete impossible.

Publicado en July 17, 2026

Building a read-only email triage agent with Nexo, the new Ruby AI agent harness, one capability at a time across a multi-post series.