chore(chain-drain): land Codex overnight chain into main #52

Merged

pdurlej merged 20 commits from codex/swarmheart-leviathan-readonly-preview into main

2026-05-10 14:14:34 +02:00

pdurlej commented

2026-05-10 14:13:56 +02:00

Owner

War-room cleanup 2026-05-10 — Prof Opus

Problem detected: Operator zmergował 17+ Codex chained PRs (#31, #33–#50) earlier today, ale każdy z nich zmergował się do swojej base branch (chain-internal), nie do main. Dlatego po uga buga merged queue jest puste, ale main NIE ma 9665 LoC chain content.

Diagnosis:

git log --merges origin/main pokazuje tylko #28, #32, #51 jako merged-to-main
Forgejo state=closed, merged=true for #31/#37/etc — TAK, ale do codex/agent-machine-identities, codex/columns-policies-and-wip etc., NIE main
Pan Herbata wczoraj ostrzegł o tej pułapce (PR base_ref trap, modelizm-#2 z 2026-05-09 letter)

Fix (this PR):

Tip branch codex/swarmheart-leviathan-readonly-preview zawiera wszystkie 19 commits chain content (after all chained merges, this is the canonical "all of overnight Codex shipping" snapshot)
Plus right-now merged origin/main into this tip (absorbs #28, #32, #51 = ESLint fix)
Now git diff origin/main...HEAD shows clean drain: ~9665 LoC, 53 files

Scope

This PR is a chain drain merge — single click brings all of Codex overnight work to main:

#31 agent attribution in card activity
#33 web agent proposals inbox
#34 cards lightweight relations
#35 mcp card relation tools
#36 columns policies + WIP
#37 mcp policy warnings
#38 web blocked policy cues
#39 manual projection drift detection
#40 AI Inbox conventions hardening
#41 MCP ops smoke + health
#42 mcporter registration smoke
#43 web machine activity audit view
#44 created-list backfill script
#45 transition history read path
#46 flow metrics v0
#47 flow health settings view
#48 leviathan projection contract
#49 leviathan readonly preview

(NOT included: #50 leviathan write apply — operator may have merged into #49's head separately, verify before this drain.)

Risk

Single big merge vs sequential chain merges. If anything fails, entire chain rolls back as one.
Per-PR review notes still NOT addressed (operator-decision per blocker — see individual PR comments).
ESLint canary cleared via #51 ✅
Some PRs flagged product/tech blockers in 3+3 review — operator merging here = "good enough" decision.

Test plan

After merge: pnpm --filter @kan/db build && pnpm --filter @kan/api typecheck && pnpm --filter @kan/api test should pass per individual PR test plans
Visual smoke: Kan-ductor mobile UI loads, AI Inbox renders, agent proposals visible
Production deploy: deferred (per Codex's "no production deploy in this PR" pattern)

Rollback

git revert -m 1 <merge_sha> rolls all chain content out of main in single commit.

Identity discipline note

Branch was Codex's, merge into chain done by operator. This PR opener pdurlej (MCP layer), but content authorship preserved via individual commits.

— Prof Opus 🦍🎓 (war-room continuation, Codex tokens out, drain attempt for chain not-on-main)

## War-room cleanup 2026-05-10 — Prof Opus **Problem detected:** Operator zmergował 17+ Codex chained PRs (#31, #33–#50) earlier today, ale każdy z nich zmergował się do swojej **base branch** (chain-internal), nie do `main`. Dlatego po `uga buga merged` queue jest puste, ale main NIE ma 9665 LoC chain content. **Diagnosis:** - `git log --merges origin/main` pokazuje tylko #28, #32, #51 jako merged-to-main - Forgejo `state=closed, merged=true` for #31/#37/etc — TAK, ale do `codex/agent-machine-identities`, `codex/columns-policies-and-wip` etc., NIE main - Pan Herbata wczoraj ostrzegł o tej pułapce (PR base_ref trap, modelizm-#2 z 2026-05-09 letter) **Fix (this PR):** - Tip branch `codex/swarmheart-leviathan-readonly-preview` zawiera **wszystkie 19 commits chain content** (after all chained merges, this is the canonical "all of overnight Codex shipping" snapshot) - Plus right-now merged `origin/main` into this tip (absorbs #28, #32, #51 = ESLint fix) - Now `git diff origin/main...HEAD` shows clean drain: ~9665 LoC, 53 files ## Scope This PR is a **chain drain merge** — single click brings all of Codex overnight work to main: - #31 agent attribution in card activity - #33 web agent proposals inbox - #34 cards lightweight relations - #35 mcp card relation tools - #36 columns policies + WIP - #37 mcp policy warnings - #38 web blocked policy cues - #39 manual projection drift detection - #40 AI Inbox conventions hardening - #41 MCP ops smoke + health - #42 mcporter registration smoke - #43 web machine activity audit view - #44 created-list backfill script - #45 transition history read path - #46 flow metrics v0 - #47 flow health settings view - #48 leviathan projection contract - #49 leviathan readonly preview (NOT included: #50 leviathan write apply — operator may have merged into #49's head separately, verify before this drain.) ## Risk - **Single big merge** vs sequential chain merges. If anything fails, entire chain rolls back as one. - **Per-PR review notes** still NOT addressed (operator-decision per blocker — see individual PR comments). - ESLint canary cleared via #51 ✅ - Some PRs flagged product/tech blockers in 3+3 review — operator merging here = "good enough" decision. ## Test plan - After merge: `pnpm --filter @kan/db build && pnpm --filter @kan/api typecheck && pnpm --filter @kan/api test` should pass per individual PR test plans - Visual smoke: Kan-ductor mobile UI loads, AI Inbox renders, agent proposals visible - Production deploy: deferred (per Codex's "no production deploy in this PR" pattern) ## Rollback `git revert -m 1 <merge_sha>` rolls all chain content out of main in single commit. ## Identity discipline note Branch was Codex's, merge into chain done by operator. This PR opener pdurlej (MCP layer), but content authorship preserved via individual commits. — Prof Opus 🦍🎓 (war-room continuation, Codex tokens out, drain attempt for chain not-on-main)

pdurlej added 20 commits

2026-05-10 14:13:56 +02:00