Apple Mail CLI addresses a practical automation problem: an instruction to prepare or send a message is not the same as evidence of what Mail contains. The project exposes local Mail operations through structured JSON, using Python and macOS JavaScript for Automation. The installed 0.4.0 implementation is shared by Codex/ChatGPT and Hermes; this note examines a separate 0.4.3 review candidate.
Reconcile the plan with Mail
Its campaign workflow compares a plan with Sent and Send Later evidence. Matching uses real Message-ID, the exact To/CC/BCC recipients, sender and schedule. Scheduled attachment checks use filename, size and byte-hash evidence from the local message where readable. Historical Sent checks explicitly retain their weaker metadata-only assurance. A conflicting copy or insufficient evidence blocks continuation instead of being hidden by one apparent match.
Bind approval to the actual intent
The 0.4.3 candidate changes direct send to a preview-first interaction. Confirmation must match the digest of the reviewed sender, ordered recipients, subject, body and attachment evidence. Changed content invalidates that confirmation. A durable local reservation is made before automation and prevents another attempt of the same intent after acceptance, timeout or an ambiguous outcome. The digest binds content; a human decision still supplies authorization.
What testing demonstrates
Independent synthetic review passed the bounded guard scenarios, including concurrent confirmations producing one backend invocation. This is evidence about protocol behavior under controlled conditions, not proof of real email delivery. Native read-only checks reached the local index and schema without launching Mail. Live account, draft, send and scheduling acceptance remain incomplete. The candidate distinguishes submission accepted by Mail from confirmed delivery.
Next validation / proposed
The review candidate was separate from the installed 0.4.0 tool. Version 0.4.3 source has since been published on GitHub as experimental; publication does not establish native acceptance or delivery. The next stage remains a separately authorized native validation with controlled sender, recipient, content and readback. Private messages and fixtures are excluded from this record. Public source and validation guidance
NLINEA · Implementation and bounded synthetic review; experimental public source
← All records