Repository navigation
docs(rfc): record one live MCP Authorize card per server - #24872
kyleseaman wants to merge 1 commit into
Conversation
UX Review (Opus 5.5) — ⏭️ skippedRevision |
First Principles Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: a docs-only change that puts a product-owner reversal on record, with its consequences and rejected alternatives written down. It takes nothing away and renders nothing new. Notes
[DESIGN-REVIEWED] 32de16b Description read[DESCRIPTION-READ] 2efdf5b3df19db9a9ce8f0894645bafad8af92413ffb344fd6ece73a6f10aca6 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 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: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Review detailsNo findings. Both candidates are true statements about the RFC's description of the current code, but the RFC's decisions and exit criteria still hold against main, so neither affects a user, an operator, data or a security boundary. Notes:
[OPUS-REVIEWED] 32de16b Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
4bbbddc to
e42024d
Compare
|
HOLD at e42024d. The RFC is accurate against main: the cited functions and the code-comment rule it replaces are where it says, and #24871 records the same product-owner decision. CI is green and the GPT/Opus lanes report no findings. One blocker: the Design Review's Notes bullet is unanswered. It asks what happens when a timed-out card is removed and no next offer arrives for that server (for example the user stops prompting), and wants Rollout exit 2 to say so, so #22907 does not decide it alone. The Consequences bullet (L47) says the card disappears "until the next offer", which assumes one comes. To clear: add one sentence to Consequences or Rollout exit 2 (L59) stating the no-offer outcome as accepted, or reply on the PR explaining why an offer always follows a live-process timeout. — crew-pr-reviewer (r129) |
e42024d to
a3ef5eb
Compare
|
Addressed the HOLD in Consequences now reads:
Rollout exit 2 adds:
No dead card is left behind in either case. |
a3ef5eb to
a7c8175
Compare
|
HOLD at a7c8175. The r129 blocker is cleared: Consequences and Rollout exit 2 now say that a removed card may have no successor and that this is accepted (L47, L59). CI is green and the other lanes report nothing. One item is still open: the Design Review on this head notes that Decision 3 (L37) follows from decisions 1 and 2. As written, it could read as a request for a second dedupe path in #22907 and #24701. To clear it, say in Decision 3 that it follows from 1-2, or reply on the PR explaining why it is a separate rule. — crew-pr-reviewer (r131) |
a68d3c3 to
ba3b50b
Compare
|
Maintainers confirmed the design direction on a68d3c3. The RFC file on this head is byte-identical to that one. The PR now waits for a maintainer approval on ba3b50b, because the force-push dismissed the earlier approval. Review result on this head: no blocking findings. The change is docs only: one RFC and one row in the request-for-change index. The code anchors the RFC cites are all still on main: Optional:
|
ba3b50b to
3e309ca
Compare
3e309ca to
32de16b
Compare
Code change approver recommendation: ApproveWhy: the decider estimates a 10% chance that this change introduces a defect, within the 12% limit for approving on its own, and every check passes. Docs-only RFC recording the product owner's decision to remove replaced and timed-out MCP Authorize cards instead of greying them out, keep the dead-process recovery card, and let a late success still turn the card green. I checked the decision and exit criteria against main: renderMcpOAuthMessage still keeps every retired card visible, the comment it replaces is quoted correctly, and the dead-process path sets no expired_reason, so decision 4 can be checked. Some context lines are stale (main already has attempt_ended and shows a timed-out card as a red error notice, not grey), but that is wording drift and does not change behaviour. No trust boundary moves; accepting an RFC on a product owner's decision matches existing index rows. An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit |
Problem / Motivation
Goal: Record the product owner's decision that a chat shows one live MCP Authorize card per server, and removes replaced and ended cards instead of greying them out, so review lanes read it from main.
The fixes for #7580 and #7654 left a rule in a code comment in
McpOAuthBanner.tsx: a retired Authorize card stays visible as a grey notice, because "a vanished banner ... is indistinguishable from the sign-in having succeeded." Kyle Seaman (product owner) decided on 2026-10-26 to reverse that. With rotating sign-in offers, the grey cards bury the one live button. The decision lives only in conversation and PR dispositions today, so the First Principles and UX lanes on #22907 block every head of it as reversing a pinned decision with no record.Why it matters
#22907 and #24701 implement the decision and cannot reach readiness while their lanes see no recorded decision. The First Principles lane reads RFC status from the base branch only, so the record has to land on main before those PRs rebase.
Not a goal
docs/decisions/entry. That directory accepts decisions from maintainers listed inMAINTAINERS.md.What changed (motivation → approach → change)
docs/request-for-change/rfc-mcp-authorize-card-one-live-per-server.md,status: acceptedon the product owner's decision of 2026-10-26 (#24871)docs/request-for-change/README.mdThe RFC states the decision: a replaced card and a timed-out card are removed from the transcript, a chat shows at most one live Authorize card per server, the dead-process recovery card from #7654 stays, and a late success still shows as the green authenticated card. It names what it replaces (the code-comment rule from #7580 and #7654), the consequences (users lose the visible history of failed attempts; the next offer mints a fresh card), and the two alternatives considered (greyed cards; a collapsed history line). It follows the shape of the other product-owner RFCs in the directory, such as
rfc-fold-recorded-edit-cards.Backwards compatibility
Compatible: documentation only; no code, config or contract changes.
Removes nothing: the diff adds one document and one index row.
Tests
Proof: not provable -- docs or config only
sh scripts/docs-lint.shpasses ("All documentation checks passed").Manual verification
N/A — docs only. Every code claim in the RFC was checked against main at
2ec42c9dc1:renderMcpOAuthMessageand itssuperseded/expiredbranches inwebsite/src/pages/chat/McpOAuthBanner.tsx,_supersede_open_mcp_oauth_bannersinsrc/kiro_crew/dashboard/chat_turn/mcp_session.py, and_expire_dead_child_oauth_metainsrc/kiro_crew/dashboard/chat_utils_display.py.Related Issues
Closes #24871
Refs #22907
Refs #24701
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)