Cryptograss:Moods/ramp/todo

• ☐
  Replace Jenkins' github-token (maybelle_github_token) with a fine-grained token: cryptograss repos Jenkins builds, Contents read-only, Commit statuses write; nothing else

  infra · justin

  Today's classic token can push to maybelle-config production and magenta main on GitHub without githouse, and deploys still clone from GitHub, so a pushed commit reaches the next deploy. Code run by Jenkins on the controller (arthel, pickipedia builds) can read it from the environment. Jenkins only reads repos and posts preview comments on justinholmes.com commits; if a fine-grained token can't comment without Contents write, drop the comment. Goes with magent's 'deploys clone from githouse' item, which closes the same hole from the other side. A person makes the token and puts it in the vault.
• ☐
  hunter's ssh router refuses a banned person's key; a redeploy prints any change in people, keys, githouse_mergers or MOOD_ADMINS (scripts/render_diff.py)

  code · magent

  Plan section 10. Not built: neither is on maybelle-config production (52dc3a2c), checked 10 Nov.
• ☐
  Merge maybelle-config:magent/jenkins-from-githouse (brings re/githouse-owns-new-names and magent/githouse-owns: all three at once), redeploy maybelle, then hunter

  merge · justin

  Justin, 17:55Z: build the Jenkins part now and merge together. 0335c24f; re to review the Jenkins half (her lane). re reviewed and fixed the flip, 17:14Z. ci/test green (githouse 56). The first hunter deploy after merging runs the old deploy-hunter.sh (from GitHub); the playbook parts take effect at once. After this, a GitHub-button merge doesn't deploy: githouse logs it, and root adopts it.
• ☐
  Jenkins' pickipedia and arthel jobs check out from githouse (a fetch-only key for Jenkins), so those deploys stop following GitHub's production

  code · magent

  After githouse-owns. With re (Jenkins is her lane).
• ☐
  The ramp shows githouse's github.json: a deploy branch GitHub has moved around githouse (ahead or apart)

  code · magent

  After githouse-owns is live. Until then it's only in githouse.log.
• ☐
  Jenkins audit, re's open items: a policy on silent skips (pickipedia node/lua checks), pickipedia's chain-data warning that deploys stale silently, NVM/pnpm fetched from the network in every pipeline

  code · re

  From re's test map, 30 Oct. The mtime guard (#244) and the alarms (#402, #275) are done.

Done (87)

• ☑
  Merge magenta:magent/branch-links (a candidate named in a message, magenta:re/foo or magenta#7, links to its page in the ramp), then redeploy maybelle

  merge · justin

  Merged as e563759, 10 Nov ~05:23Z; live after maybelle's redeploy. It didn't tick itself: only GitHub pull request links did (fixed by the item below).
• ☑
  Merge magenta:magent/todo-branch-merged (an item linking a candidate branch's page ticks when githouse merges or closes it; a closed branch answers its ask), then redeploy maybelle

  merge · justin

  98d9cd8, from main at e563759. Full suite 1939 OK. No migrations. Justin's ask, #ramp 05:24Z. Once live, this item ticks itself when it merges: the first real test.
• ☑
  Redeploy maybelle: tonight's magenta main (8eb880e: merged-and-bar, re/tiding-torrent-test-url) and maybelle-config production (52dc3a2c: branches-agent-init) aren't live

  deploy · justin

  Redeployed by Justin ~05:5xZ. Checked 06:0xZ: live is main at fac415d (the wordmark, branch-links and todo-branch-merged are all on the page). Correction: magent's 05:15Z 'not live' check was invalid; it fetched the page signed out, and the merge bar's code is sent only to someone who can merge.
• ☑
  Merge magenta:magent/ask-anywhere (an ask tag on a line of its own counts, with words after it), then redeploy maybelle

  merge · justin

  e2a33a1, from main at fac415d. Justin, #ramp 06:41Z: magent's merge ask showed as text, a goodnight under the tag. Full suite 1943 OK. No migrations. Independent of tick-summary; one redeploy covers both.
• ☑
  Merge magenta:magent/tick-summary (closed is final, kept like merged; the wiki bot's tick says githouse, not only GitHub), then redeploy maybelle

  merge · justin

  412cab7, from main at fac415d. re's review point (05:42Z). Full suite 1939 OK. No migrations. Small: it can wait for the next maybelle redeploy.
• ☑
  Redeploy hunter: magenta:re/candidates-news changes the poller

  deploy · justin

  On main since 73f095d (checked 10 Nov 04:48Z). Poller only, so hunter, not maybelle. Redeployed: hunter's containers were recreated 05:59-06:00Z, and the poller's mounted source has the candidates-changed code (checked 06:0xZ).
• ☑
  Merge magenta:magent/merge-dots (merge asks as one row of small pills, dots and the branch; the sheet leads with title and description; opening a page says nothing in the Mood), then redeploy maybelle

  merge · justin

  d2b8e93: main (b1d112c, with mood-list-candidates; its ramp.py conflict resolved) and magent/ramp-drops-merged merged in. Justin's asks, #ramp 14:10Z. Full suite 1953 OK. No migrations. Checked in a browser on a local server with githouse's real copy (pills, and the sheet for re/help-install-app).
• ☑
  Merge magenta:magent/ramp-drops-merged (a merge pressed in a Mood marks the ramp's board old the moment it's merged), with merge-dots, which includes it

  merge · justin

  d61545d. Justin, #ramp 14:13Z: mood-list-candidates, pressed in #interfaces, stayed on the ramp until he left it and came back. The test fails without the fix. Ticks itself when merge-dots merges.
• ☑
  Merge magenta:magent/deploy-stamp (the ramp's 'live on' learns what each deploy put on each server: magenta on hunter, at last)

  merge · justin

  92d77d1, from main at 562e74f. Full suite 1948 OK. No migrations. re to review. Either order with the maybelle-config half: each ignores what the other doesn't send yet.
• ☑
  Merge maybelle-config:magent/deploy-commits (hunter's and maybelle's playbooks note the magenta commit they deployed; report-deploy.sh sends it when the deploy ends)

  merge · justin

  1afc8906, from production at 52dc3a2c. ci/test green (playbooks parse, every task's options check); report-deploy.sh's new tests post to a listener of their own. re to review: it's deploy logic.
• ☑
  Redeploy maybelle (magent/ask-anywhere, magent/tick-summary; then deploy-stamp and deploy-commits once merged), then hunter, so the ramp first says 'live on hunter'

  deploy · justin

  ask-anywhere and tick-summary merged 06:52Z and read 'not yet' on the live ramp (checked 08:1xZ). hunter after maybelle: its report is the first to carry magenta's commit. Done by Justin ~15:57Z: the ramp's 'live on' read 'hunter, maybelle' for everything merged that day.
• ☑
  Merge maybelle-config:magent/repackage-cid-check (/enrich/repackage takes a CID and nothing else), before delivery-kid's redeploy

  merge · justin

  1f02a4c5, from production. re's hardening point, 16:26Z. Its test and the 5 repackage tests pass. One redeploy then covers both.
• ☑
  Redeploy maybelle (magent/merge-github-only-branch) and delivery-kid (magent/resdtp-releases, and repackage-cid-check once merged)

  deploy · justin

  maybelle done: redeployed 16:39Z, magenta 925cd58 (/api/servers/, checked 16:4xZ). delivery-kid and the 26 repackages moved to Cryptograss:Moods/uploads-and-embeds/todo (Justin, 16:42Z: that work lives there).
• ☑
  Merge magenta:magent/merge-github-only-branch (the merge sheet finds a branch pushed only to GitHub by its pull request: 18 open were 'githouse has no …'), then redeploy maybelle

  merge · justin

  159f5fe, from main. Justin, #ramp 16:10Z. Full suite 1955 OK; the new test fails without it. No migrations. githouse already merges such branches (it reads GitHub's head).
• ☑
  Merge maybelle-config:magent/resdtp-releases (#298: repackage the 26 older release videos so they play on iPhones), then redeploy delivery-kid

  merge · justin

  Still open, checked 10 Nov 16:0xZ; delivery-kid runs 52dc3a2c (redeployed 07:12Z), which has the standard-range fix for new videos but not this. Skyler's iPhone is the test.
• ☑
  Redeploy delivery-kid: it runs 9fd62b2d (5 Nov), so the iPhone fix merged 9 Nov 17:39Z (strip-empty-sdtp) isn't live

  deploy · justin

  Checked /version 10 Nov 04:48Z. Redeployed 07:12Z: /version says 52dc3a2c (checked 16:0xZ), so strip-empty-sdtp is live. #290 and #298 weren't in it; #298 is its own item above.
• ☑
  Verify Cryptograss:Moods/Instructions (revs 15842-15849): merge items link their branch page, deploy items don't; merge and redeploy asks name repo:branch, from any turn; no GitHub pull requests, push and ask by name

  verify · justin

  Before it, the page (rev 14232, 4 Nov) said asks come only from look-only turns and named 'which PR'. Every emergent reads this page on every wake, so this is what teaches the other Moods the ramp.
• ☑
  Verify Cryptograss:Moods/ramp/Plan: six revisions of magent's since your last verify (rev 14836, 6 Nov), in sections 8, 10 and 11

  verify · justin

  Section 8 (merging without a PR, as built), 10 (rev 15035: one tier, phone merges everywhere; no merge can take away your way to stop it), 11 (branch tests on Jenkins, after re's review). Then 10 Nov 08:03-08:05Z (revs 15808-15818, Justin's ask at 08:03Z): stale lines corrected in sections 1-3, and a dated 'Built' or 'As of 10 Nov' line in 4, 6, 8, 9, 10 and 11, so the page says what's running. One verify covers all of it.
• ☑
  Verify Cryptograss:Runbooks/A key is compromised: never verified; all 8 revisions are magent's

  verify · justin

  Newest is 15094 (8 Nov 02:15Z: the brakes and AZ5 live). Its freeze brake is rev 15038.
• ☑
  Verify re's candidates-changed and candidates-same lines on Cryptograss:Emergents/Wake (rev 15582)

  verify · justin

  re's edit, 9 Nov 17:54Z, after your 17:42Z verify. Until verified, the poller sends the same words from poller/wake_texts.md.
• ☑
  Delete magenta's dead production branch on GitHub (a634f16, Nov 2025), or say it stays

  decision · justin

  Still there, checked 10 Nov 05:15Z. It's why the ramp measured every magenta branch with no pull request against a year of history (+85,583 lines). The ramp routes around it now; deleting it removes the trap for anything else that guesses production first.
• ☑
  Check which account maybelle_github_token (Jenkins' github-token) belongs to and its scopes, without printing it: pickipedia and arthel merges run code in Jenkins with it

  infra · justin

  If it can administer the org, a merge there can change the org and its deploy keys (Plan, section 10, open).
• ☑
  hunter's disk (82%, warning at 85%): decide whether hunter prunes unused images, which the playbook leaves alone on purpose

  decision · justin

  Punted, Justin 07:15Z. Checked 10 Nov 06:5xZ. The housekeeping role says unused images are 'deliberately left alone ... That is a human decision.' magent's 30.1 GB figure was docker's 'reclaimable', which counts layers shared with images in use (magenta:latest is 10.6 GB and runs four containers). Tagged images no container uses come to ~7 GB: playwright/mcp 1.4 GB (10 months old), php:8.2.35-apache twice (715 MB each), groovy:4.0-jdk17 707 MB, memory-lane-source-watcher 639 MB, then many small ones. If yes, magent writes the maybelle-config change: a weekly 'docker image prune -a' for images unused a week or more, beside the build-cache prune.
• ☑
  A deploy watch (server, commit): deploys stamp the commit they deployed, and the Mood is told

  code · magent

  Plan, section 4. The merge half is done (#415; #444 says githouse's merges in their Mood). Built 10 Nov 08:2xZ as magent/deploy-stamp and maybelle-config:magent/deploy-commits (above). Found while building: the ramp already knew magenta on maybelle and maybelle-config everywhere; hunter's magenta was the gap. Not yet: a redeploy ask resolving on the commit rather than on any finished deploy; pickipedia's code, which Jenkins deploys.
• ☑
  Deploys clone from githouse's copy on maybelle instead of GitHub (Jenkinsfiles, deploy scripts)

  code · magent

  re's point 9. Justin's go, 10 Nov 16:54Z. Built as maybelle-config:magent/githouse-owns (36a265eb), the flip too: githouse had taken the deploy branches from GitHub every minute, so cloning from it alone closed nothing. Jenkins' pickipedia and arthel jobs are the item below.
• ☑
  Review maybelle-config:magent/githouse-owns: githouse owns main/production (GitHub a mirror; ahead or apart there is never taken; root's adopt); maybelle, hunter and the deploy scripts fetch from githouse

  review · re

  Look hardest at reconcile() and _merge(): can anything still move a deploy branch here from GitHub without root's adopt? And the cold-start fallbacks: can anything but a missing githouse trigger them?
• ☑
  Review the reworked maybelle-config#250 (githouse on maybelle, mirror on hunter; your ten points of 2 Nov folded in)

  review · re

  Reviewed 6 Nov 04:44Z: sound, two bugs found (the hooks lacked the deploy key; the port check was inverted), both fixed on the branch by magent from re's description.
• ☑
  Merge maybelle-config#250 and redeploy maybelle

  merge · justin

  Merged 05:01Z (07cb4d10); maybelle redeployed 05:08Z, failed=0, every githouse task ran. githouse's sshd answers on 10.0.0.2:2222, key only, and nothing on maybelle's public address.
• ☑
  Redeploy hunter for #250: the mirror, and the containers' keys

  deploy · justin

  Done 05:22Z. The relay worked (keys on maybelle, authorized_keys written), but sshd refused them all: the git account was locked. Fix below.
• ☑
  Merge maybelle-config#280 (githouse: the git account was locked) and redeploy maybelle (then hunter at its next deploy)

  merge · justin

  useradd left the git account locked ('!'); sshd refuses a locked account even by key. Reproduced and fixed in a throwaway container on hunter, 6 Nov 05:35Z. After maybelle's redeploy, the mirror fills within a minute and pushes work.
• ☑
  Confirm jenkins_github_ssh_key can push to the org's repos: after hunter's redeploy, re pushes a branch to hunter and magent reads githouse.log on maybelle for a forward failure

  test · magent

  6 Nov 05:42Z: yes. magent pushed magenta:magent/githouse-smoke to maybelle from its container; the branch was on GitHub within seconds and in hunter's mirror within a minute.
• ☑
  Merge magenta#408 (PR cards look at GitHub hourly), then redeploy maybelle

  merge · justin

  Merged 04:57Z and live (magenta at 3a57cb2a).
• ☑
  maybelle-config#246 (read-only GitHub token for memory-lane): merge or close

  decision · justin

  Optional now that #408 is live. Justin, 6 Nov: no appetite for another token. Close it unless the asks starve again. Closed on GitHub 7 Nov 00:40Z.
• ☑
  Re-verify Cryptograss:Moods/ramp/Plan: section 5 rewritten per your 6 Nov decisions (main stays default, merge commits only, no rulesets)

  verify · justin

  Verified 6 Nov 11:22Z at rev 14724; section 5 rewritten 19:57Z; re-verified by Justin 22:49Z (rev 14836).
• ☑
  Set each repo's GitHub default branch to the branch that deploys

  infra · justin

  Dropped, Justin 6 Nov 11:22Z: main stays the default, the branch where history is final; the deploy branch stays per-repo config (#376). The mains catch up to the productions after the ramp settles: arthel and maybelle-config are fast-forwards, pickipedia has 2 commits on main not on production, magenta's production is dead (3 commits, Nov 2025).
• ☑
  Merge commits only (squash and rebase off) on pickipedia-mcp, the one of the six still allowing both

  infra · justin

  Done. 7 Nov 00:39Z, read back with gh api: all six repos githouse copies are merge-commit only; squash and rebase off everywhere.
• ☑
  Rulesets on the branches that deploy (PR required, no force-push, no deletion; admins bypass)

  infra · justin

  Dropped, Justin 6 Nov 11:22Z: no further entrenching GitHub. githouse's check() already refuses any push that isn't <name>/*, so no emergent reaches a deploy branch; at the flip GitHub becomes a mirror we push to, and a ruleset is one more thing to undo. Plan section 5 to be rewritten once item 2 is decided.
• ☑
  Merge magenta#333 (repo:branch autocomplete, reading /api/branches/ from maybelle's githouse copy), then redeploy maybelle

  merge · justin

  Reviewed by re 1 Nov and 6 Nov. Its mount exists on maybelle since the 05:08Z deploy, so it can go now.
• ☑
  Build the ramp view on magenta: open PRs across our repos, last five merged, who and where, last two mentions, sorted by last push or last mention, temperature

  code · magent

  Built as /moods/pulls/ in magenta#420, 7 Nov 08:00Z. Plan, section 2.
• ☑
  Review magenta#420, the ramp view (/moods/pulls/)

  review · re

  A fix is a re/ branch of your own; magent merges it in (Plan, section 3).
• ☑
  Merge magenta#420 and redeploy maybelle

  merge · justin

  After re's review.
• ☑
  Review magenta#421: the ramp page's mentions from an index kept up from new posts (on maybelle #420's query ran past its 4 s budget)

  review · re
• ☑
  Merge magenta#421 and redeploy maybelle

  merge · justin

  After re's review. Until then /moods/pulls/ works, without mentions.
• ☑
  Merge magenta#423 (~ opens every mergeable in the composer) and redeploy maybelle

  merge · justin
• ☑
  Verify the githouse practices in Cryptograss:Emergents/GeneralOrders (rev 14966)

  verify · justin

  Verified 9 Nov 17:44Z (rev 15579).
• ☑
  Decide Plan sections 8 and 9 (merging without a PR; a page per piece of work): verify, or push back

  decision · justin

  re reviews the design first. Open questions: SSH signature per merge (CLI, then passkey on the phone) versus trusting a signed-in device; who may merge. Settled by building: merges through githouse since 8 Nov, a page per work since #448; section 10 answered both questions (one tier, githouse_mergers justin and skyler). What's left is the Plan's verify, above.
• ☑
  Review maybelle-config#286: githouse merge, a branch merged on an order signed with a person's SSH key (Plan, section 8)

  review · re
• ☑
  Merge maybelle-config#286 and redeploy maybelle

  merge · justin

  re reviewed it 17:37Z; her notes are fixed in 1ca7f8c7. Since then one commit, 53a17642: githouse_mergers is [justin, skyler] (Justin, 7 Nov 23:53Z), with a test for skyler's RSA key.
• ☑
  githouse brakes: refuse orders by a name banned in memory-lane, and every order during AZ5 or a root-only freeze (magenta.sh freeze); merge nothing while memory-lane doesn't answer

  code · magent

  Built 8 Nov: githouse's half on maybelle-config#286 (61e88fe8), memory-lane's /api/brakes/ on magent/githouse-brakes. The freeze is /mnt/persist/githouse/etc/frozen, root only; magenta.sh freeze as a wrapper is still to do.
• ☑
  Review the brakes: maybelle-config#286 since 1ca7f8c7 (61e88fe8, 53a17642) and magenta#426 (/api/brakes/)

  review · re

  Reviewed 8 Nov 00:28Z: two fixes (an unknown name answered 'not banned'; a malformed answer escaped brake()), merged in by magent as 2b4289f7 and 113c488.
• ☑
  Merge magenta#426 and maybelle-config#286, then redeploy maybelle

  merge · justin

  Merged and deployed 8 Nov 02:14Z.
• ☑
  magenta.sh merge: fetch the head itself, show git log and the diffstat it computed (tested: never text a service hands it), sign in magenta-merge, hand to githouse over ssh to maybelle

  code · magent

  Built 8 Nov as magenta#428 (stacked on #426), sent through memory-lane as a mailbox rather than ssh; githouse fetches orders in maybelle-config#287 (stacked on #286).
• ☑
  Review the mailbox and magenta.sh merge: magenta#428 and maybelle-config#287

  review · re

  Reviewed 8 Nov 01:29Z: two bugs, fixed by re in re/merge-mailbox-fixes (merged in, 47d1176). Her open point, the signature screen at the mailbox with a limit per signer, was done by magent in 9b61b18.
• ☑
  Merge magenta#428 and maybelle-config#287, then redeploy maybelle

  merge · justin

  Merged and deployed 8 Nov 02:14Z. Checked from hunter at 02:15Z: /api/brakes/ answers justin false, skyler false, an unknown name null; the mailbox is empty; /whose/ names both mergers' keys.
• ☑
  The first merge by mail: magenta.sh merge magenta:magent/mergeable-tilde (#423), from an updated magenta checkout

  test · justin

  Done 8 Nov 03:23Z with #429 instead (#423 went by GitHub's button): 8dd57b26 on main, committed by githouse, 'Signed by justin (githouse merge order 5a83f9084734)', its parents main's tip and exactly #429's head e9f8e601. GitHub marked #429 merged, by cryptograss-maybelle.
• ☑
  Review the merge button: magenta#433 and maybelle-config#289

  review · re

  Reviewed 8 Nov 16:56Z: the server side holds; one race in the sheet, fixed by re (re/merge-sheet-race, merged in as a618edd). magent added a guard against opening a sheet while a press is in flight (326744b) and re's memory-lane name assert (1122f499).
• ☑
  Merge magenta#433 and maybelle-config#289, redeploy maybelle, then press Merge… on something real

  merge · justin

  Merged and deployed 9 Nov 00:34Z. First press 00:38Z: 162b767 on magenta main, 'Merge magenta:journals-section … Pressed by justin, signed by memory-lane'. Follow-up: one green Merged button, magenta#442.
• ☑
  Review Plan section 11: branch tests on Jenkins without its secrets (an agent for branches, credentials scoped to trusted jobs)

  review · re

  Reviewed 9 Nov 00:49Z: the branch's own Jenkinsfile could choose the built-in node, so branch jobs run a fixed pipeline from maybelle-config instead; Authorize Project; the whole environment, env_file included. Folded in, rev 15304.
• ☑
  Decide where the branches agent runs: a separate small box (re's and magent's pick) or maybelle behind an internal network and egress proxy

  decision · justin

  Justin, 9 Nov 01:00Z: on maybelle; he'll pentest it, and it moves if a real hole turns up.
• ☑
  Review branch tests: maybelle-config#292 (the agent, its network and proxy, the fixed job, githouse's trigger) and magenta#443 (ci/test)

  review · re

  Reviewed 9 Nov 01:18Z: the agent's secret was readable by ci/test (fixed: ci/test runs as builder through run-ci, 9ed34798); a venv per build, IPv6 off, more proxy checks (same commit); re/ci-test-fixes merged into #443 (acb7aaa). Her item 4, narrowing the reporter, goes with the rest of section 11.
• ☑
  Merge maybelle-config#292 and magenta#443, then redeploy maybelle

  merge · justin

  Merged 9 Nov; the deploy at 01:35Z failed in a second, at playbook load (an apostrophe in the isolation check's heredoc), so nothing ran. Fix: maybelle-config#293.
• ☑
  Merge maybelle-config#293 (the isolation check as a file; ci/test checks playbooks parse) and redeploy maybelle

  merge · justin

  Merged; the 01:45Z deploy got halfway and failed on mem_limit (docker_container says memory). Fix: #294.
• ☑
  Merge maybelle-config#294 (memory, not mem_limit; ci/test checks every task's options) and redeploy maybelle

  merge · justin

  Merged and deployed 01:56Z: the agent is online, the isolation check passed. branch-tests was on disk but never loaded (a startup race); fix: #295.
• ☑
  Merge maybelle-config#295 (reload Configuration as Code after Jenkins starts; check every job exists) and redeploy maybelle

  merge · justin

  Merged and deployed. 9 Nov 02:47Z, end to end: magent pushed maybelle-config:magent/branch-tests-smoke to githouse; githouse asked Jenkins; branch-tests #1 ran on the branches agent through run-ci, every ci/test section green, SUCCESS.
• ☑
  Section 11, the rest: the reporter's Build narrowed to branch-tests and the deploy-report jobs (first: its password is on githouse now); credentials into a folder only the trusted jobs see; the controller's environment emptied (env_file included)

  code · magent

  Reporter narrowed: maybelle-config#296, merged and deployed 9 Nov 10:51Z. Folder move dropped (Justin, 10:58Z). Emptying the environment, and the trusted agent it needs to mean much, moved to #security (Justin, 11:07Z).
• ☑
  Merge magenta#448 (a page for each piece of work on Tidings, with its diff and link-to-line; contains #446, the 👍) and redeploy maybelle

  merge · justin

  Syntax colours added (3547f52, Justin's ask). re reviewed 15:09Z: nothing blocks; her binary-file 500 fixed in acec0dd. Branch test #9 passed at acec0dd. Two migrations (0050, 0051). Merging #448 merges #446 with it.
• ☑
  Merge magenta#447 (the emergents' activity bar on Tidings and the ramp) and redeploy maybelle

  merge · justin

  Branch test #6 passed. Independent of #448; one redeploy covers both.
• ☑
  Review magenta#449: work comments whole in the Mood and Mood replies on the page; one merged line; ramp names link to pages; ask dots; a merge sheet per Mood

  review · re

  Reviewed 15:59Z: both places safe; one bug (a refused or expired press hid GitHub's merge line), fixed by re in re/merge-line-refused, merged in as cf51b50. 1861 tests OK.
• ☑
  Merge magenta#449 and redeploy maybelle

  merge · justin

  After re's review. No migrations. Full suite 1859 OK locally.
• ☑
  Merge magenta#451 (candidate branches on every wake; the ramp says candidate branch; stacked on #449), then redeploy maybelle and hunter

  merge · justin

  #449 merged 16:37Z. Now also: merge asks one line per candidate with each asker's dot, and answered asks leave the page without a reload (038bdb2). Full suite 1868 OK. No migrations.
• ☑
  Verify the 'candidates' line and the two 'which candidate branch' edits on Cryptograss:Emergents/Wake

  verify · justin

  Until verified, the poller sends the same words from poller/wake_texts.md. Verified 9 Nov 17:42Z (rev 15578).
• ☑
  Merge magenta:magent/ramp-see-the-diff (every candidate named and linked alike; ± See the diff on each row; the diff first on its page; sizes against main, not magenta's dead production), then redeploy maybelle

  merge · justin

  Merged at c9ab117 (with the merge sheet's ± See the full diff) and live by 19:10Z: the ramp's rows read → main.
• ☑
  Merge magenta:magent/sheet-diff-fast (a work's page reads the last board: no ~8 s wait after ± See the full diff), then redeploy maybelle

  merge · justin

  Merged as a8197ee.
• ☑
  Merge magenta#452 (sections at /ramp/, /tidings/…; now with main merged in and a candidate's page at /ramp/<repo>/<branch>/, Back returning where it came from), then redeploy maybelle

  merge · justin

  0804ab7: also the ramp sorted by Mood and a button per person (only theirs / theirs hidden), Justin's ask 19:47Z. Full suite 1896 OK. No migrations.
• ☑
  Merge magenta:magent/ramp-blocks (pushed in blocks ago; the ramp's column heads sort it), then redeploy maybelle

  merge · justin

  Merged and live: on 10 Nov 01:19Z all 51 open rows read in blocks, and the four heads sort.
• ☑
  Merge magenta:magent/merge-bar (Merge now: the sheet goes, a bar below the nav with ✓ or ✗; the ramp drops what merged), then redeploy maybelle

  merge · justin

  Merged as 07a6f53 and deployed by 10 Nov 01:17Z.
• ☑
  Close magenta#442 (the sheet's green Merged button): superseded by magent/merge-bar

  decision · justin

  Closed; magent deleted its branch on githouse, 10 Nov 04:5xZ, so it leaves the ramp.
• ☑
  Merge magenta:magent/merged-and-bar ('Merged lately' lists every branch githouse merged; the merge bar on every page), then redeploy maybelle

  merge · justin

  Merged as 8eb880e, 10 Nov ~04:53Z; live after maybelle's next redeploy.
• ☑
  FIRST: merge magenta:re/tiding-torrent-test-url. main fails two ReleaseTorrentTest tests since #452 moved tidings to /tidings/, so every magenta branch tests red until it merges

  merge · justin

  Checked 10 Nov 01:20Z: both fail on main (07a6f53); branch-tests #31 passed on re's fix. Tests only, nothing to redeploy. Merged by Justin, 10 Nov ~04:53Z.
• ☑
  Merge maybelle-config:magent/branches-agent-init (the branches agent gets an init as PID 1; an aborted build no longer jams the next), then redeploy maybelle

  merge · justin

  Merged 10 Nov 04:31Z (52dc3a2c); live after the next maybelle redeploy. cfd176c0, with production merged in. Second half (10 Nov): #34 was aborted, run-ci's cleanup never ran, its workspace stayed builder's, and #35-#36 failed at deleteDir. run-ci now cleans up when stopped (a trap, ci/test waited for in the background), and the job runs 'run-ci reclaim' before deleteDir; tried in a throwaway image on hunter: a TERM mid-test cleaned up in 0 s, a jammed workspace reclaimed and deleted. First half: 80925401. Branch tests #19-#21 failed on 'Resource temporarily unavailable': the agent's java was PID 1 and reaped nothing, so orphaned git and sh stayed <defunct> until all 2048 pids were held (Justin's ps, #ramp 18:59Z). The 19:10Z redeploy cleared them for now; this stops them piling up again. Its own branch test, #22, passed.
• ☑
  Merge magenta:re/candidates-news (the whole candidate list on a session's first wake in a Mood, then only what changed), then redeploy hunter

  merge · justin

  bfa95f6b, from main. Full suite 1873 OK. No migrations; poller only, so hunter, not maybelle. Merged (on main at 73f095d, checked 10 Nov 04:48Z); hunter's redeploy still owed.
• ☑
  Merge magenta:re/test-cache-own (test runs get a cache of their own; the test_moods flakes)

  merge · justin

  2e4488c6. Tests only, nothing to redeploy. Merged (on main at 73f095d, checked 10 Nov 04:48Z).
• ☑
  Merge magenta#442 (the sheet's green Merged button) and redeploy maybelle

  merge · justin
• ☑
  memory-lane tells the Moods what githouse merged, from /srv/git/.merges (read-only mount)

  code · magent

  Built 9 Nov as magenta#444 (the Mood an order came from gets the merged line).
• ☑
  Merge maybelle-config#296 (the reporter starts only branch-tests and deploy-hunter), magenta#444 (merges by mail said in their Mood) and magenta#445 (red and green on the ramp), then redeploy maybelle

  merge · justin

  All three passed their branch tests on Jenkins (#2, #3; #445 is #4 when githouse sends it).
• ☑
  Review magenta#415: merge asks and to-do ticks read merged from githouse's copy (refs/pull/N/head ancestry; refs/merged/ for a repo:branch), GitHub only as fallback

  review · re

  Reviewed 7 Nov 00:55Z: two real findings (an unguarded subprocess; an ask naming an unknown repo could clear on its known half). re pushed the fixes as re/fix-asks-and-githouse (05e515c), her first commit through githouse; magent merged it with one correction (her clearing condition stood every ask). 1741 tests pass.
• ☑
  Merge magenta#415 and redeploy maybelle

  merge · justin

  After re's review.