You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
SKILL amendment: issues live where the fix lands — narrow the repo:* transfer lane to seam cards, adopt file-at-destination filing #7165
The maintainer questioned the central-backlog design —「我还是不懂,既然每次分诊也要看 objectui 的仓库,为什么相关的 issue 不能直接提交到 objectui 仓库? cloud也是。」— and then ruled「同意」on the PM's proposal:
Execution cards live in the repo where the fix lands (file-at-destination).
Exception A: the maintainer keeps filing everything into objectstack without pre-routing; the triage seat transfers such cards out to the target repo instead of only labeling them.
Exception B: seam cards stay in objectstack — cards whose content IS the cross-repo ordering (contract-first chains: a spec/contract change with downstream objectui/cloud adaptation, console-bump linkage chores, anything whose Blocked-by ordering across repos is the substance).
Rationale accepted with the ruling: the original reasons for "one main backlog" have partly expired — objectui and cloud now run their own whole-repo seats and real local backlogs (127 / 47 open at the 2026-08-10 snapshot), and #6903 already establishes "exactly one producer per backlog". The double-track (a repo:objectui lane of ~47 cards in objectstack beside objectui's 127 local cards) produced the v17 producer gap, double bookkeeping, and two-query release inventory.
Task: amend .claude/skills/pm-dispatch/SKILL.md
Multi-repo coordination rule 1 ("One main backlog"): rewrite to "issues live where the fix lands", with exceptions A and B above stated verbatim as the two allowed deviations.
Filing discipline (os-dev definition + Prime Directive 10 filing rules): a cross-repo discovery made while working in one repo is filed directly into the repo where the fix lands, with a body backlink to the discovery context — no repo:* transfer card for pure sibling-repo fixes.
Triage seat charter (rule 4 bijection): for maintainer-filed cards that land in objectui/cloud, the routing action becomes TRANSFER to the target repo (GitHub issue transfer, same org — thread preserved, old URL redirects), not a repo:* label. The seat remains the only producer of routing decisions on the objectstack backlog.
repo:objectui / repo:cloud label semantics: narrowed to seam cards only. Do not delete the labels; update their descriptions.
The one-time migration of the existing transfer-lane stock is a separate execution card (filed alongside this one) and does NOT wait for this SKILL PR — its authority is the maintainer ruling quoted above; this card is the durable record.
Guardrails
This is a .claude/ protocol change: the PR needs review by another seat or a maintainer walkthrough — no self-merge.
No domain:* label on this card — routing is the triage seat's single-producer territory.
Routing: domain:devx applied directly by the PM session.
Maintainer standing directive (this session, 2026-08-10):「以后你应该直接路由」, followed for this card by「7165 你直接设定路由吧」. Lane precedent: #6903, the previous pm-dispatch SKILL amendment card, was routed domain:devx.
Scope addition for the SKILL PR this card produces — record the standing directive itself: the maintainer-facing PM session routes cards directly at filing/triage time (applies domain:* now, and transfer-based routing once this card's rule-1 rewrite lands), quoting the authorization in the audit comment. The "no domain label, wait for the triage seat" deferral is retired for maintainer-directed filings. The triage seat remains the producer for the organic backlog sweep; the body's closing line ("No domain:* label applied") predates this directive and is superseded by this comment.
Maintainer ruling (2026-08-10)
The maintainer questioned the central-backlog design —「我还是不懂,既然每次分诊也要看 objectui 的仓库,为什么相关的 issue 不能直接提交到 objectui 仓库? cloud也是。」— and then ruled「同意」on the PM's proposal:
Rationale accepted with the ruling: the original reasons for "one main backlog" have partly expired — objectui and cloud now run their own whole-repo seats and real local backlogs (127 / 47 open at the 2026-08-10 snapshot), and #6903 already establishes "exactly one producer per backlog". The double-track (a repo:objectui lane of ~47 cards in objectstack beside objectui's 127 local cards) produced the v17 producer gap, double bookkeeping, and two-query release inventory.
Task: amend .claude/skills/pm-dispatch/SKILL.md
Coordination
Guardrails