1.9.0a5 — co looks finished, and the first run shows your people

An opt-in preview of 1.9.0. Stable is 1.8.9.

One look for every co command (#1997)

• One palette. Commands you can run are cyan, counts bold, paths dim, warnings yellow, errors red, and every result ends on a Next: line. It comes from one module (connectonion/cli/style.py), so the same thing looks the same wherever it appears.
• Plain when it should be plain. Colour follows the terminal: NO_COLOR, TERM=dumb, pipes, logs and launchd get plain text, and --json never changes. The words are the same either way; only the styling differs.
• co audit checks it. Each page now runs twice, once as an agent reads it and once as a terminal shows it. A page fails the new look rule when it has colour in a pipe, no colour in a terminal, or different words in the two runs. The help-contract test in CI fails on any such page, and the list of pages allowed to fail it is empty.
• Found by the second run: co commands, co proxy --help and all fifteen co rem help pages printed no colour at all. co doctor cut long paths to an ellipsis at 100 columns. All are fixed.

co rem you can read (#1996)

• co rem status is a dashboard, not a field dump:

  • the state and the next run;
  • pages written of mapped, per category, skills included, and the next page to write;
  • today's runs, pages changed and tokens;
  • each mailbox with ✓ or ✗ and the command that fixes it;
  • the week's budget and the last run in one line.

  The internal fields are behind --verbose, and --json is unchanged. Next is the step the first line names: init, then start, then logs once the schedule runs. Before, it always said logs.

• co rem init shows progress. Each stage it can count (mail windows, bodies, session files, skills) gets a bar, and each model turn a spinner with its time so far. It ends by repeating the notebook's counts, which had scrolled away. Off a terminal it prints exactly the lines it printed before.

A first run that shows the people around you

co rem init used to write your own page with the pre-#1850 quick pass (about 15k tokens of sampled mail) and no one else's page. Now, in a terminal, after saying what it will spend:

1. Your whole page: investigate me, from everything you sent and your coding sessions of the last 30 days (or --days), in one model turn over evidence files. About 15 minutes.
2. The 3 people you wrote to most in the last 14 days, one turn each.
3. The projects you worked on in the last 14 days, one call each.

The whole run stops starting pages at 5 points of the Codex week, half the notebook's weekly 10, or at the 70% floor, whichever comes first. Ctrl-C keeps the map and every page written. co rem investigate people writes more people, and investigate me --quick is still there by hand.

Fixed from running it on real data (#2008)

Running this preview on the owner's real mail and sessions before tagging it found the first run would bill over 10M input tokens while promising ~90k per page. Fixed before release:

• A first run that says what it will cost. Capped at your page, 3 people and 3 projects (--first-people, --first-projects to change), people read the run's own window, and one line before anything is spent gives the pages, the billed input tokens and the minutes, estimated from this notebook's own runs. Measured: announced 1.4M, spent 1.25M (#2011).
• Your own page has its own shape. It says who you are and what you are working on from your coding sessions; a role is yours only when you say it or others say it of you (it had taken a partner's roles). Your name comes from how people address you, not the mailbox's display name (#2011).
• A written page has no "not investigated yet" left in it, and a project page draws its overview when the material shows the architecture (#2011).
• An upgraded notebook is tidied at its next map or sync, with nothing deleted: services off the people list (22 on the owner's notebook), your own addresses folded into your page, duplicate skills merged, old coverage lines removed; --mine never offers a real person (#2010, #1999).
• The reader and status count from one census: "with findings" means written, times are local, an interrupted run shows at once, and Next is what "To write next" names (#2010).
• One look, checked for what a terminal shows: no [env] lines on every command, co status on the palette with a Next line, inbox checks without stray colour, co doctor the same in a terminal and a pipe; co audit's look rule gives stderr a terminal too and fails repeated lines, words coloured in pieces and status output without a Next line. 302 of 302 pages pass (#2009).
• A maintenance turn stays under 15k whatever the source, and no turn is pointed at the rationale docs (#2006); urllib3 2.8.0 for three new advisories (#2007).

Install

python -m pip install --upgrade 'connectonion==1.9.0a5'
co rem status
co audit co rem

Known limits

• The first-run estimate comes from one notebook's runs; a new machine uses the owner's measured defaults until it has its own.
• The docs site and O Chat still say "wiki"; they live in other repositories.
• Sharing from a page (DD-074, #1968) is designed, not built.