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
decide: who watches upstream releases, and how often #184
Parent: #158. Split out of #163 so the policy could land without this blocking it.
#163 settled four of its five required decisions: mechanism, review checklist, attribution, and what we never take. docs/UPSTREAM.md carries them.
It deliberately did not settle cadence, and said so in the document:
Not settled by this document. Who reads upstream releases, and how often, is an owner and team decision. Until it is decided, upstream is read on demand — when someone wants a specific fix, or when a release note happens to mention something that matters to us.
Why this needs a decision rather than staying open
The adversarial review of #163 pushed back on leaving it open, and the pushback is fair: a policy's job is to create accountability where there was none, and declaring the status quo as policy guarantees the failure it warns about.
What "on demand" costs, stated plainly: an upstream security fix or crash fix can sit unnoticed until it bites us. Upstream runs about 88 commits per 30 days against 4 releases[api: repos/Kpa-clawbot/CoreScope, 2026-10-07], so there is a steady stream to not be watching.
The counter-argument, which is also real: a named obligation nobody performs is worse than an honest gap. The imported agent surface sat untouched for four months while appearing maintained (#144).
What has to be decided
Who reads upstream releases. A person, not "the team".
How often. A minimum frequency, not "regularly".
Where the "worth taking" decision is recorded when the answer is no. task: how we track upstream after the fork link is gone #163's policy says a rejected change gets one line on an issue; this should confirm that is the mechanism, or replace it.
A reasonable shape to argue with, not a recommendation: whoever is doing CoreScope work that month checks gh release list -R Kpa-clawbot/CoreScope when they start, which is cheap and self-scheduling, with an explicit accepted risk that quiet months go unwatched.
Not in scope
The mechanism for taking a change is already decided and documented. This issue is only who looks, and how often.
Done when
docs/UPSTREAM.md's "Cadence and ownership" section names a person and a minimum frequency, replacing the "not settled" text, and the owner has agreed to both.
Activity