Mission: restore the umbrella structure (#9919–#9931) and file all unorganized
open issues under it. Triage/organize only — nothing was fixed.
Predecessor: 2026-06-11 triage (PR #9932, UMBRELLA_PLAN.md).
| Umbrella | Found | Closing cause | Action |
|---|---|---|---|
| U1 #9919 | CLOSED (state_reason: completed) | Manual close by agent 2026-06-11T19:14Z — timeline closed event has no commit and no PR source; closed at batch-session end after the last member PR (#9948) merged |
REOPENED — theme has open issues (#9825 upstream-blocked, #9949/#9950 prose-noted but never task-listed, #10026 new) |
| U2 #9920 | CLOSED (state_reason: completed) | Manual close by agent 2026-06-11T18:05Z — same pattern, after PRs #9988/#9989 merged | REOPENED — theme has 5 open follow-ups (#9980, #10001, #10002, #10026, #10027), four cross-referenced onto the issue after it closed |
| U3–U13 #9921–#9931 | open | — | no action needed |
- Neither close came from a PR closing keyword — both were end-of-session agent
closes on task-list completion. The lifecycle rule (human-only closes) is now
codified in
UMBRELLA_PLAN.md§ Umbrella lifecycle. - Reopen comment posted on both: "Reopened: umbrella epics close only by explicit human decision, not by PR keywords or task-list completion."
- Labels: all 13 umbrellas still carry
umbrella— none lost it, none re-added. - Prevention sweep: 3 open PRs (#10004, #10028, #10029) — zero closing keywords aimed at #9919–#9931; no PR body edits were needed.
- 151 open issues; 13 are the umbrellas themselves.
- Organized set: 124 issues already in umbrella task lists — including 5 post-triage issues that wave sessions had already appended (#10019/#10022/#10023 → U6, #10024 → U9, #10025 → U4).
- Work set: 42 issues, all created 2026-06-11 (wave-review follow-ups and live-console findings). No pre-#9942 issue was missed by the first triage pass.
Closed stale: 0. Every candidate was checked against the 22 PRs merged since
2026-06-11T12:00Z — all 42 were produced by those PRs' reviews (follow-ups),
none is resolved by them. Duplicates merged: 0 (near-pairs #9965/#9999,
#9982/#9996, #10011/#10012 verified as distinct scopes).
Unfiled / needs-taxonomy-decision: 0 — all 42 placed. Filed: 42, appended
to umbrella bodies as dated ### Triage delta sections (existing lists preserved):
| Umbrella | Filed | Issues |
|---|---|---|
| U1 #9919 deployment-install | 17 | #9949 #9950 #9956 #9965 #9966 #9967 #9982 #9996 #10000 #10005 #10006 #10007 #10010 #10016 #10020 #10021 #10026 |
| U2 #9920 data-layer | 4 | #9980 #10001 #10002 #10027 |
| U3 #9921 code-intelligence | 1 | #10008 |
| U4 #9922 api-contract | 8 | #9958 #9959 #9983 #9985 #9986 #10011 #10012 #10013 |
| U5 #9923 llc-module | 5 | #9951 #9978 #9987 #9992 #9995 |
| U7 #9925 transcriber | 1 | #9968 |
| U11 #9929 product-features | 6 | #9942 #9943 #9984 #9999 #10017 #10018 |
Boundary calls worth review:
- #10001 → U2, #10026 → U1 per the dispatch boundary rule; the ordering edge (#10026 must land before #10001) is noted inline on both entries — #10001's fix would start running the repaired chain against never-migrated populated DBs.
- #9968 → U7 (not U4): the IDOR regression was introduced by U4's PR #9955, but the transcriber umbrella owns the broken auth surface (subsystem-owner rule).
- #9951 → U5 (not U1): #9919's prose listed it as a "new member", but the
defect lives in
llc/scheduler/heartbeat_scheduler.py— subsystem owner is U5. - #10016 (SLM lifecycle completeness) is itself an epic; filed as a sub-epic inside U1 rather than inventing a 14th umbrella (taxonomy changes are the human's call — promote it if you prefer).
- #10010 → U1: LLC API surface, but the defect class is single_user Postgres gating policy (#9765/#9913 class), which U1 owns.
Dependency edges recorded inline on both ends: #10026⇄#10001 (ordering), #9984 blocks #10018 (nav gating before rebuild), #9999⇄#9965 (UX facet vs deploy-side cause), #10027 prerequisite-for #10026's guard option.
Labels applied: agents → #9951/#9978/#9987/#9995; infrastructure →
#9966/#9967/#10010/#10021; needs-human-decision → #9983 (reimplement-or-retire
is a product call). bug was already present on every defect-shaped issue.
- #10035 (→ U9 #9927) — this session's freshly-pushed unmerged remote branch
was deleted within ~2 minutes of push (before the PR could be created); the
scheduled
branch-cleanup.ymlis exonerated (last run 06:22 UTC, pre-push). Suspect: parallel-session merged-branch remote pruning misclassifying the branch. Issue includes the guard spec for #9917's implementation. The session's first local worktree was also swept by a parallel session's start-protocol (commit-less branch == "merged" — by-design; mitigation: anchor-commit immediately after worktree creation).
Co-located / deploy-update surface — 9 of 42 new issues trace to it: #9965 #9966 #9967 #9982 #9996 #10006 #10007 #10020 #10021. Suspect recently-merged PRs: #9991 (one-click update, → #9996 #10021), #9933/#9952 (co-located nginx routing, → #9966 #9967), plus latent never-tested paths (#9982 resolve_drift, #10020 dual autobot_shared). This matches the 2026-06-11 bug-wave finding that the co-located/deploy surface has no automated gate — #10023 (U6, co-located smoke gate) is the structural fix and should be treated as that cluster's exit criterion.
Secondary cluster: LLC test-harness trust (#9987 #9995 + adapter blind spots
#9992 #9951) — suspect area is llc/tests full-suite isolation, not any single PR.
- U1 #10026 → U2 #10001 (strict order, release-gating)
- U4 live-console contract cluster (8 issues, user-visible breakage)
- U1 deploy-surface cluster + U6 #10023 gate (the regression cluster above)
- U7 #9968 (IDOR, priority-high — jump the queue)
- U11 #9984 → #10018/#10017 rebuild track
Full rationale in
UMBRELLA_PLAN.md§ Dispatch-order delta — 2026-06-12.
LICENSE/NOTICE/SPDX untouched (read-only per mission). No code changed.