Repository navigation
Manager card: no briefing limit tied to the merge queue (owner ruling 2026-10-26) - #2534
Conversation
Owner ruling 2026-10-26: a work-in-progress limit on briefing is never part of the process. Builders and Managers work at full speed, Managers batch into combined PRs, a merge queue forms, and the Lander works it down. - docs/roles/manager.card.md, "What this seat owns": add "Brief at full speed" with the ruling. - "What it must not do": remove "Brief more than the Lander can land". - "On arrival" step 4: keep finding the Lander, drop sizing a wave on the open PR count. No other text in docs/roles, docs/METHOD.md, CLAUDE.md or .claude limits briefing by queue depth or merge rate; no test or script pins the removed phrases. The card stays under tests/test_role_cards.py's 6 KiB cap (6142 bytes with CRLF). Proposed PR title: Manager card: no briefing limit tied to the merge queue (owner ruling 2026-10-26)
Review round 1 (round-af70715a): "the queue" alone was narrower than the ruling and could read as the merge queue only, leaving the open PR count, the metric the removed On-arrival step used, still usable as a throttle. Name all three inputs. Drop the restated batching sentence: the wave paragraph above already says it, with its own-PR exceptions, so stating it again unconditionally contradicted them. Card is 6134 bytes (CRLF), under the 6144 cap. The korus roles/MANAGER.md throttle text is outside this repository and is being corrected by a separate Builder.
Review round 2 (round-166a7e2a): - The card is a summary of CLAUDE.md section 5, and the ruling lived only in the card. Add it to the Manager row of the seat table so the source the card derives from carries it. - The three-input list read as complete. Head the rule "No work-in-progress limit on briefing" and say "at least" the open PR count, the merge queue and the merge rate never limit it. - Drop "the Lander works the queue down", which the Lander row states. Card is 6134 bytes with CRLF, under the 6144 cap.
|
Companion korus PR: MEFORORG/korus#246. Land both so the card and the playbook agree. |
Owner ruling 2026-10-26, in session: the five-item cut on combined PRs is removed. A wave PR is cut when every Builder in it has reported, or before the Manager closes its instance. The own-PR exceptions stay: an item that fixes a red main, changes a security control, supersedes an ADR, or must land in order against another open PR gets its own PR, and a red or conflicting item goes back to a Builder. The card states it in the no-work-in-progress paragraph, which already carries the 2026-10-26 ruling, and the CLAUDE.md Manager row gains the same few words. A grep of docs/roles, CLAUDE.md, docs/METHOD.md, .claude/ and tests/ found no engine text carrying a four-item Builder lane ceiling or a seat-owned burn start-throttle, so nothing else moved. PR title: Manager card: brief at full speed, no work-in-progress limit
Code review round 1 on the five-item cut. The no-work-in-progress text read as lifting every limit on briefing, and CLAUDE.md wins over korus, so a Manager could brief through a Steward hold. The owner ruling of 2026-10-26 keeps usage limits with the Steward's holds and the 90 percent Workflow gate, and both the card and the CLAUDE.md Manager row now say so. "No item count caps a PR" read against the own-PR exceptions, so it now names the wave PR. The card is held under the 6144-byte cap by tightening two sentences whose content stays: the wave-branch merge step and the commit-without- asking line. Left for the Manager: scripts/coord/lane.ps1 still defaults -Capacity to 4, outside this brief's grep scope; korus MANAGER.md still carries the five-item cut until the korus companion change merges.
Code review round 2. The previous commit said the 90 percent Workflow gate stops a new brief, which overstates a gate that covers starting a Workflow, and it copied a live threshold into two always-loaded files. Both now say usage limits still bind and point at korus STEWARD.md (and COMMON.md in CLAUDE.md), naming no number. "Never limit it" now says "briefing". The shared commit-authority sentence the previous commit shortened is restored to match the other cards, and On-arrival step 4 is dropped: it checked nothing, and the must-not list already hands each PR to the Lander. The card is 6126 bytes on a CRLF checkout.
|
Follow-up pushed: f69d66b..908ac2b (three commits). Owner ruling 2026-10-26 ("Remove the cap"): the five-item trigger is gone from the Manager card's wave paragraph. A wave PR is cut when every Builder has reported, or before the instance closes. The own-PR exceptions stay unchanged. The card's no-work-in-progress paragraph and the CLAUDE.md Manager row now say no item count caps a wave PR, and that usage limits still bind (a Steward hold, the Workflow gate), pointing at korus rather than restating a threshold. A grep of docs/roles, CLAUDE.md, docs/METHOD.md, .claude/ and tests/ found no other engine text carrying a five-item cut, a four-item Builder lane ceiling, or a seat-owned burn start-throttle. Open for the Manager:
Card size: 6126 bytes on a CRLF checkout (cap 6144). QA: pytest tests/test_role_cards.py, test_coord_seat_roster_verdict.py, test_claude_section_citations.py, test_claude_md_wiki_state_root.py, test_coord_seat_prompt.py: 148 passed. code-review at xhigh, two rounds (round-5e682854, round-d16c1e07). Tag (korus review_tag.py, exit 0 for both rounds, same line each): |
|
Lander: review proof now covers korus 246 at A Lander-dispatched
These two land together once their rows are on vault |
|
Lander: enqueuing at pinned head |
github-actions
Bot
added
the
ci-red
|
CI failed in Its own head can still be green: the queue revalidates the merge, and the path gates that skip on a pull request run there. Read the run before retrying. https://github.com/MEFORORG/MessageFoundry/actions/runs/40408417109 |
Records the owner ruling of 2026-10-26: flow management never includes a work-in-progress limit on briefing. In the owner's words: "This is NEVER a thing in our processes. Builders and managers work at full speed. Managers batch work into combined PRs as much as possible. A merge queue develops. The lander works it down."
Changes:
docs/roles/manager.card.md: removes the "Brief more than the Lander can land" must-not; adds a "No work-in-progress limit on briefing" paragraph under What this seat owns; On-arrival step 4 no longer sizes a wave on the open PR count. Card is 6134 of 6144 bytes in a CRLF checkout.CLAUDE.mdsection 5, Manager row: carries the same ruling, since the card says it is derived from that table.Merge note: korus
roles/MANAGER.mdstill carries the "when to brief NOTHING" throttle. Its companion correction is korus branchmanager-no-wip-limit(PR to follow). Until both land, the card and the playbook disagree.Follow-ups, not filed here:
scripts/coord/pr_state.py(PR_PAGE=100) andscripts/ci/check_stalled_prs.py(LIST_LIMIT=300).Checks: pytest
test_role_cards,test_coord_seat_roster_verdict,test_precompact_reprime_hook,test_enter_worktree_check,test_afk_kill_switch(294 passed); thentest_role_cards,test_claude_md_wiki_state_root,test_claude_section_citations,test_review_standards,test_sds_rule_ids_are_stable,test_coord_seat_roster_verdict,test_ci_tooling_gate(194 passed). Pre-commit passed on all three commits. No Python changed, so no ruff or mypy.QA: code-review at level xhigh, two rounds. round-af70715a tag:
xhigh effort -> 10 inline angles -> dedup (no verify) -> sweep -> <=15 findings; round-166a7e2a tag:xhigh effort -> 10 inline angles -> dedup (no verify) -> sweep -> <=15 findings. Both read with korusscripts/quality/review_tag.py(exit 0, printed with --ascii). Findings fixed in 4edf5b6 and f69d66b; the unfixed ones are the follow-ups above.