[BrewingCoder]

OP2-284, Gmail subtask 7 of 7: the live probe, and the README. Everything in #8 to #11 was written against Google's documentation and documented response shapes. This PR adds the test that checks it against a real account, and updates the docs to describe what was built, honestly.

The README marks Gmail "built against the Gmail API documentation; live verification pending." Once LiveGmailProbe has passed on your machine, change that row to done. The ticket's done-when is "no claim the probe hasn't verified", and it hasn't run yet.

LiveGmailProbe

Alongside the other Live* probes, and skipping (not failing) unless the environment is there:

• the dev Vault and database;
• a Google OAuth client in ~/.claude/credentials.json, as outpost2_google_client_id and outpost2_google_client_secret;
• a Gmail connection made through the app. OP2_PROBE_LABEL picks one by name.

It has three tests, in increasing order of consequence.

1. Reading. Runs every time.

   • A cold access token from the stored refresh token (OP2-279's refresh path).
   • The inbox, newest first, with real received times.
   • A full read whose body has no HTML or   left in it.
   • Two search pages with nothing repeated, the second resumed with a different query that must be ignored.
   • Labels as folders with counts, including Inbox and Trash, and without Starred, Unread or Important.
   • An attachment's bytes opened twice and compared. Gmail issues a new attachment id on every fetch, and opening by part id has to survive that.

2. Organising, undone. Runs every time. On the newest inbox message:

   • star, then restore;
   • toggle read state, then restore;
   • make a temporary label Outpost probe <timestamp> and apply it as a category, checking the message is still in the Inbox;
   • move the message to that label, checking it's out of the Inbox, then move it back.

   In a finally, the label comes off and is deleted, and the message's labels are checked identical to before. Every check reads Gmail's own record of the labels (format=minimal), not what the tools claim.

3. Sending. Opt-in with OP2_PROBE_SEND=1.

   • Sends high-importance mail to the account itself and waits for it to arrive. Importance must read back as high, and the body must survive.
   • Replies, and checks that Re: isn't doubled and that both messages share one Gmail thread.
   • Moves the message to Trash and back.
   • Then moves the probe's own messages to Trash.

   It never mails anybody else. It's opt-in because a test that sends every time it runs is a test that gets disabled.

README

• "58 tools across four providers", and the design note now covers routing mail by provider.
• What Gmail serves and how it differs: labels not folders (move, archive, categorise, delete-no-mail), Trash and Spam, stars without due dates, importance shown in Outlook but not in Gmail, no sender blocking, and times in UTC.
• Setting up the Google OAuth client, and Testing mode: only listed test users can connect, and sign-ins expire after seven days. The README says the resulting "reconnect" is that limit, not a fault (OP2-283).
• Running the probe, including the opt-in for sending.
• Status: Gmail as live verification pending; Gmail for users outside a test list as blocked on OP2-283; Atlassian and Azure DevOps still planned.
• Gotchas: Gmail's documented quirks (offline access and consent for refresh tokens, per-fetch attachment ids, page-token scope, maxResults=0 rejected), flagged as documented and not yet hit live.
• The test count is updated to 1,076.

Testing

All three probe tests skip here, as designed: there's no dev Vault in this environment. They can only really run against your account. Full suite: 1,076 tests; only the 20 Vault-dependent web-host tests fail, as on main in this environment.

To run it locally

dotnet test --filter "FullyQualifiedName~LiveGmailProbe"
OP2_PROBE_SEND=1 dotnet test --filter "FullyQualifiedName~LiveGmailProbe"

The first command runs reading and organising; the second adds sending. Running them with --logger "console;verbosity=detailed" prints what was seen at each step.

---

Generated by Claude Code