Repository navigation
docs(rfc): screen 03 becomes a startup prune of sync-generated crewmates - #13032
Conversation
|
Intent: record in the accepted CrewMates launch RFC the decision the product owner took for screen 03 on 2026-09-23 — a one-time startup prune of sync-generated crewmates instead of a user-facing opt-in — so the First Principles lane reads the real product shape off main for #12798. |
Design Review (Fable 5) — ✅ PASSDesign-level review of Design-Verdict: PASS Verify the cited 2026-09-23 product-owner decision is actually on record — the RFC amendment carries no link or approver for it. Suggestions
[DESIGN-REVIEWED] 46ae3ed |
Opus 5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
…ed crewmates The CrewMates launch RFC decided screen 03 as a user-facing opt-in step. The product owner replaced that on 2026-09-23: crewmates an earlier sync generated are pruned once at gateway startup (never-chatted rows removed, chatted rows untouched on their V1 binding), with no UI; a person with custom agents and no crewmate is served by screen 02's empty state and New crewmate. Only the screen 03 section, its backward-compatibility bullet, its acceptance bullet and the #12798 entry in the PR list change.
005d137 to
46ae3ed
Compare
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Problem / Motivation
docs/request-for-change/rfc-crewmates-launch.md(accepted) decides screen 03 asa user-facing opt-in step: "lists the existing custom agents, pre-checked, with
two actions: Add N crewmates and Not now". The product owner replaced
that decision on 2026-09-23, and the implementing PR
(#12798) now ships a
startup migration with no UI. The First Principles lane reads product shape off
the base branch, so the RFC on main must say what was actually decided.
Why it matters
Until the RFC records the new decision, #12798 is blocked as a product-shape
change without an RFC, and the launch's decision record describes a screen
nobody is building.
What changed (motivation → approach → change)
Docs only, one file. The screen 03 section is replaced by the decision taken:
member_id, shareddefaultstore, bound to the person's own custom agent) that were never chatted with; chatted ones stay exactly as they are on their V1 binding; only the config row goes, agent files and transcripts stay; unreadable history removes nothing; a marker makes it run onceOne short paragraph records the reasoning: the generated rows are leftover
state, not a choice; a memory binding is identity and is not rewritten; no
screen to maintain; the removal drops nothing the person made, one click
re-enrols any agent. It also says what happens to a generated row the person
edited but never chatted with (removed with its edits; accepted, because the
row is the sync's and the crewmate is one click away) and that member-aware
edits stamping
member_idtake a row out of the rule.Three references to the old step are updated so the RFC does not contradict
itself: the § 6 backward-compatibility bullet (config key → marker file), the
§ 7 acceptance bullet (now names exactly the rows § 03 removes — package-sourced
rows and chatted rows stay), and the #12798 entry in § 8. Every other line
of the RFC is byte-identical. The section names #12224 as the change that
removed the enrol-on-mount call the generated rows came from.
Backwards compatibility
Compatible: a documentation change.
Tests
N/A — docs only;
scripts/docs_lint.pypasses.Manual verification
N/A — docs only.
Related Issues
no linked issue: amends the accepted launch RFC for #12798.
Checklist
feat|fix|docs|refactor|perf|test|chore|ci|build|revert: ...)