Skip to content

feat(dashboard): show when a PR's checks are still running - #21357

Merged
iamwhatever merged 1 commit into
mainfrom
feat/pr-checks-running-clarity
Oct 21, 2026
Merged

iamwhatever merged 1 commit into
mainfrom
feat/pr-checks-running-clarity

Conversation

@kyleseaman

@kyleseaman kyleseaman commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: While a pull request's checks are still running, the dashboard says so — in the sidebar chip, the Checks tab and the merge banner — even after one check has already failed.

On a PR with 101 checks where one has failed and one is still running, the sidebar chip shows only a red ×, the same as a run that has finished. The Checks tab reads Checks running 99/101: the fraction counts a failed check as "not complete", so it cannot say whether two checks are still running or one failed and one is running. The merge banner says "Branch protection requirements (approving reviews or required checks) are not yet satisfied", even when the only thing in the way is checks that have not finished.

Why it matters

Anyone driving a PR from the dashboard has to open GitHub to learn whether CI is still moving or has stopped. A red × that hides running checks reads as "done, failed", so people act on a run that is not over yet.

Not a goal

  • Changing how GitLab reports CI: its pipeline aggregate already reads running until the pipeline ends.
  • Changing the check list's order or grouping inside the Checks tab.
  • Changing the counts shown once no check is running.
  • Animating the sidebar chip: motion on a session card stays reserved for agent activity.

What changed (motivation → approach → change)

Failure wins the CI rollup, so the status the sidebar reads (ci: failed) could not tell "still running" from "finished". The gateway now projects one more chip-status field, ciPending: "1", when a rollup is failed and a check is still in flight. The PR panel's checkInFlight also leaves out a GitLab MANUAL gate, which waits on a person, when it counts running checks. Every writer of the chip cache (the GitHub and Bitbucket Server chip reads and the full-payload write-through) sets it through one shared function, a degraded read carries it forward together with ci. It is projected for GitHub and Bitbucket Server only, the providers whose chip read computes it; a third-party provider plugin's chip status stays ci and state, so its two reads keep agreeing.

Surface Before After
Sidebar chip red × red × plus the static amber pending dot while checks still run, with the tooltip Checks failed; others still running
Checks tab, a check failed while others run Checks running 99/101 Checks 1 running · 1 failed (with nothing failed it keeps Checks running 7/10)
Merge banner (GitHub blocked), a check failed while others run generic branch-protection copy red Checks failed; others still running. 1 failed and 1 still running, of 101 checks. Merging stays blocked until required checks pass and any required reviews are approved.
Merge banner (GitHub blocked), checks running, none failed Merge blocked. + branch-protection copy Merge blocked. + Checks are still running. Merging unblocks once required checks pass and any required reviews are approved. (a manual gate alone keeps the branch-protection copy)

GitLab and other providers keep the generic blocked banner: GitLab reports a merge-gating running pipeline as unstable, so its blocked means something else blocks the MR. ciPending rides the same allowlists as ci (state_parts/source_links.py projects it onto source links, ws_event_scope.py strips it for app tokens). On the client it is typed in types/index.chat.ts, types/index.sources.ts and chat-sidebar/types.ts, parsed by utils/pullRequestStatusDelta.ts, patched with ci by store/dashboardSlice.ts from hooks/websocket/serverState.ts, and rendered by chat-sidebar/SessionSourceChips.tsx; components/PullRequestPanel.tsx reads the loaded check list directly. The chip-status shapes in docs/system-specs/modules/learn-cron-dashboard.md list ciPending. New copy is in en.manual.json and every shipped translation under website/src/i18n/locales/.

Backwards compatibility

Compatible: ciPending is a new optional field. An older client ignores it and keeps rendering ci exactly as before; an older gateway never sends it, so a newer client renders today's glyphs.

Removes nothing: no field, key or endpoint is removed or tightened; the Checks tab swaps its complete/total fraction for explicit counts only while a check has failed and others still run.

Tests

  • test/test_chip_ci_pending.py: the marker is set only for a failed rollup with a check still pending; the chip read and the full payload project it identically; GitLab's aggregate never gets it; a degraded read carries it with ci; it is on both wire allowlists.
  • test/test_source_providers_chip_status.py: the full-payload projection's failed-with-pending case keeps its expectation (no provider named, so no marker); its comment now says so.
  • test/test_source_provider_plugin.py: a third-party plugin's chip read never carries the marker, so an older plugin stays steady while a check has failed and others run.
  • test/test_source_providers_bitbucket_server.py: the Bitbucket Server chip read marks a failed-plus-in-progress build like the full payload.
  • website/src/test/PullRequestPanel.test.tsx: the tab says 1 running · 1 failed; the tab keeps Checks running N/M with nothing failed; the blocked banner leads with the failure when one has failed, says checks are still running otherwise, and keeps the branch-protection copy for a manual gate and for a GitLab MR.
  • website/src/test/ChatSidebar.sourceLinkChip.test.tsx: the chip shows × plus a static dot while checks run, with a tooltip matching its accessible name, and a bare × once they finish.
  • website/src/test/pullRequestStatusDelta.test.ts, website/src/test/dashboardSlice.sourceLinkCiPending.test.ts: a delta keeps the marker only beside failed, and a CI delta without it clears the sidebar chip's marker.

Determinism: the new backend file passed 20 runs at -n0 and 10 at -n 4 --dist loadgroup --max-worker-restart=2 beside two sibling files; it and test_source_providers.py (703 ids) passed shuffled with seed=26744; test_source_provider_plugin.py passed 20 runs at -n0. The four changed frontend files passed 10 repeats and vitest --sequence.shuffle --sequence.seed=26744. Flake-pattern ratchet and CI-pinned semgrep (1.78.0, diff-scoped) are clean.

Manual verification

Ran website/scripts/capture-pr-checks-still-running.mjs against the built SPA with stubbed APIs, in light and dark themes. Each scene is asserted on the rendered page before its frame is written:

  • failing: 101 checks, 99 passed, 1 failed, 1 running, mergeStateStatus: blocked. Asserts the red banner with both counts, Checks 1 running · 1 failed, and the chip's × plus dot.
  • waiting: 101 checks, 100 passed, 1 running, nothing failed, blocked. Asserts Merge blocked. with Checks are still running and the tab's Checks running 100/101.

Screenshots / video

A check failed while others run (101 checks): sidebar chip #7001 shows × plus the amber dot (#7000, a finished failure, a bare ×), the banner is red with both counts, the Checks tab reads 1 running · 1 failed.

Failing scene, light: chip × plus dot, red banner Checks failed; others still running, Checks 1 running · 1 failed

Checks still running with nothing failed (101 checks): amber Merge blocked. Checks are still running. …, the tab keeps Checks running 100/101, the chip shows only the amber dot.

Waiting scene, light: amber Merge blocked banner saying checks are still running, Checks running 100/101

Dark theme

Failing scene, dark

Waiting scene, dark

Related Issues

None: no tracking issue was filed for this change, at the author's request.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@kyleseaman
kyleseaman requested review from a team and KiroCrewCaptain as code owners October 16, 2026 19:17
@kyleseaman
kyleseaman requested a review from CrysisDeu October 16, 2026 19:17
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Oct 16, 2026
@github-actions

github-actions Bot commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — ✅ PASS

Design-level review of f5ac587e599b92f3082e481ec2bc2217b75c2546 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

Nothing blocks: ciPending is an optional field added beside ci, all three chip-cache writers set it through one function (_record_ci), and it rides both existing allowlists.

Notes

  • When a degraded read keeps the previous ci: failed, ciPending: "1" (via carry_known_ci), the chip can keep showing "others still running" after the checks have actually finished, until the next full read. The tooltip states that as current, so it can be wrong for a while; this fixes itself on the next read.

[DESIGN-REVIEWED] f5ac587

Description read

[DESCRIPTION-READ] a690fea6f904251305144a1ac093575be23b5ca4f61b37f42ead661ced781f77

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 .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ✅ PASS

UX-level review of f5ac587e599b92f3082e481ec2bc2217b75c2546 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

Nothing blocks. All three changed surfaces (chip, banner, Checks tab) appear in light and dark screenshots, and the blind reader read each one correctly.

Notes

  • The failed_running chip dot is shrunk with scale-75, and in light theme the reader "nearly read it as 'good'" ("looks olive or green next to the red ✕"). Make the text-warn dot full size, or give it a darker amber token in light theme.
  • The Checks tab counts two different ways for the same 101 checks: 1 running · 1 failed in one state, 100/101 in the other. The reader "can't tell whether 100 of the 101 are finished or 100 are still running." Use the explicit form in the running-only state too: 1 running, or 100 of 101 done.
  • The banner detail carries the decision-critical counts but sits under a vague title. Lead the red title with the counts, e.g. "1 check failed, 1 still running".

[UX-REVIEWED] f5ac587

Description read

[DESCRIPTION-READ] 18bf204adaca4ee329cbde1523740232f95d514cad587cb75019e6c5cb4a28da

That sha256 names everything this verdict read: the PR description, plus the 7 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 .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ✅ PASS

Premise-level review of f5ac587e599b92f3082e481ec2bc2217b75c2546 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

First-Principles-Verdict: PASS

Every item traces back to the reproduced 101-check case: the chip, the tab and the banner could not tell "failed, still running" apart from "failed, finished". No item duplicates an existing mechanism.

What this change ships

Inventory (9 items) — 9 justified

Intent: let someone driving a PR from the dashboard see that CI is still running after a check has failed, without opening GitHub. This is an ADDITION.

  1. The sidebar chip shows a red × plus a static amber dot, with a tooltip, while checks still run after a failure — justified
  2. The Checks tab reads 1 running · 1 failed instead of 99/101 when a check has failed and others still run — justified
  3. GitHub's blocked merge banner turns red and gives the failed and running counts — justified
  4. GitHub's blocked banner says "Checks are still running" when nothing has failed — justified
  5. A GitLab MANUAL gate is no longer counted as running in the panel's counts or banner — justified
  6. A new optional chip-status field ciPending goes on source links and WS deltas, and is stripped for app tokens — justified
  7. A degraded chip read carries ciPending forward together with ci — justified
  8. A ciPending returned by a third-party plugin's chip read is dropped — justified
  9. The chip-status spec lists ciPending, the new strings are added to every locale, and a screenshot capture script is added (793 sibling capture-*.mjs scripts exist) — justified

[FIRST-PRINCIPLES-REVIEWED] f5ac587

Description read

[DESCRIPTION-READ] 445e0c94cc1931962eacdeabccc00dd82e7a3eede47e261b91bd2ae773a6a815

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed f5ac587e599b92f3082e481ec2bc2217b75c2546 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] f5ac587

Verdict parsed from the review's SHA-scoped output markers for commit f5ac587e599b92f3082e481ec2bc2217b75c2546.

False positive or not applicable? A repository writer can comment:
/ai-review override fable f5ac587e599b92f3082e481ec2bc2217b75c2546: <one-sentence reason>

@github-actions

github-actions Bot commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of f5ac587e599b92f3082e481ec2bc2217b75c2546 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] f5ac587

False positive or not applicable? A repository writer can comment:
/ai-review override gpt f5ac587e599b92f3082e481ec2bc2217b75c2546: <one-sentence reason>

@kyleseaman
kyleseaman force-pushed the feat/pr-checks-running-clarity branch from ebd8a36 to 92e25d3 Compare October 16, 2026 19:45
@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: the blocked banner said to wait after a check had failed (span=ac222adfe7a9)

fixed in 92e25d3: with blocked, a failed check and a check in flight, pullRequestMergeBlocker now returns tone danger, title Checks failed, detail 1 failed and 1 still running, of 101 checks. before any waiting copy.
Waiting on checks now shows only when checks run and none has failed.
test: PullRequestPanel.test.tsx failed-plus-running banner case (danger, both counts).

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: "still running" had three definitions (span=0f96fa4eaa19)

fixed in 92e25d3: one rule, a check is in flight when bucketed pending and its status is not MANUAL; backend projection._check_in_flight (used by the single ciPending writer _record_ci, now fed check rows) and frontend checkInFlight (used by pullRequestCiStillRunning/selectedSourceStatus, the banner and the tab count).
GitLab is handled like the backend: a failed job plus a required manual job gets no marker on any surface.
tests: test_chip_ci_pending.py manual-gate rows through _record_ci/status_from_full_payload; PullRequestPanel.test.tsx manual-gate case through selectedSourceStatus, banner and tab.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: the Checks tab changed with nothing failed (span=3c0354c03c86)

fixed in 92e25d3 by removing the no-failure branch: with checks running and nothing failed the tab is main's Checks running 7/10 again, and the checks_count_running key is deleted from every catalog.
The tab shows Checks 1 running · 1 failed only while a check has failed and another is in flight, the case the description names.
tests: the two main expectations /Checks running 0\/1/ are restored unchanged.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: the banner left out the failure (span=e7b0e2ebe801)

fixed in 92e25d3: the failed-plus-running blocked banner is red, titled Checks failed, and its detail names both counts (1 failed and 1 still running, of 101 checks. Merging stays blocked until required checks pass.), so it no longer sits calm beside a red tab.
evidence: re-captured light/dark screenshots in the description.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: "1 of 101" read two ways (span=94f931890c21)

fixed in 92e25d3: the waiting detail reads 1 of 101 checks still running. in English, with the same order in every shipped translation.
The failed-plus-running case uses its own wording (1 failed and 1 still running, of 101 checks).

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: the red × on the chip looks like a button (span=1e7ba5a11a5d)

The × glyph predates this PR; this PR adds the static dot beside it and gives the pair one accessible name, Checks failed; others still running.
A tooltip on the glyph group is the opposite of the chip's design: the chip is one anchor whose own title carries the PR URL and the modifier-click escape hatch, and SessionSourceChips.tsx keeps child titles off it on purpose because a child title shadows that tooltip for the region it covers.
Restyling the × is a change to the existing chip vocabulary, outside this PR's goal.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: pluralize the chip label as "1 check failed" (span=58fa6a21d195)

The chip has no count to put there: its status is the rollup (ci) plus the ciPending marker, never per-check numbers, so "1 check failed" would be invented.
The label names the state the glyphs show; the counts live in the Checks tab and the banner, which do read 1 running · 1 failed.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: GitLab failed plus manual job got the still-running marker (span=5c1b03e7c295)

fixed in 92e25d3: pullRequestCiStillRunning now uses checkInFlight (pending and not MANUAL), the same rule as the backend's _check_in_flight, so a failed job beside a required manual job sets no ciPending and the tab strip shows a bare failure.
test: PullRequestPanel.test.tsx manual-gate selectedSourceStatus case.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Oct 16, 2026
@kyleseaman
kyleseaman force-pushed the feat/pr-checks-running-clarity branch from 92e25d3 to d227c9a Compare October 16, 2026 20:09
@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: the banner title contradicted its detail (span=0712584f6f7f)

fixed in d227c9a: the failed-plus-running blocked banner is titled Checks failed; others still running. above 1 failed and 1 still running, of 101 checks., so the title no longer says the run ended.
test: PullRequestPanel.test.tsx asserts that title; evidence: the failing frames in the description.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: retitle the still-running banner so it holds both facts (span=2eace4080045)

fixed in d227c9a with the existing checks_failed_others_still_running string (Checks failed; others still running), the same words the chip and the tab-strip glyph already carry, rather than a new near-duplicate key.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • fixed: a sighted user could not confirm what the chip's dot means (span=872d3f375315)

fixed in d227c9a: the chip's icon-only failed_running group carries title equal to its accessible name, Checks failed; others still running, following the file's own precedent for the icon-only merged glyph.
test: ChatSidebar.sourceLinkChip.test.tsx asserts the group's title equals its aria-label.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: the same state looks different on each surface (span=ac30b18aff60)

The two glyphs differ on purpose and in one direction only: motion on a session card is reserved for agent activity, so pending CI on the sidebar chip is a static amber dot (pinned by the existing test renders pending CI as a static glyph, not a spinner), while inside the PR panel pending CI has always been the spinning Loader.
This PR does not add either vocabulary; it keeps each surface's pending glyph and puts it beside the failure. The dot's meaning is now stated: the chip group carries the tooltip and accessible name Checks failed; others still running.
Unifying the two glyphs would either animate session cards or freeze the panel, a design change outside this PR's goal.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • answered: Waiting on checks when blocked is really a missing approval while a non-required check runs

The panel cannot tell required from non-required checks: the payload it reads carries each check's bucket and status, not GitHub's per-PR isRequired, so no branch here can pick out required ones without a new provider read.
The banner says what is true either way: the title names what is visibly in flight, and the detail reads Merging unblocks once required checks pass and any required reviews are approved., naming the approval path too.
Once no check is running, the banner falls back to the generic branch-protection copy, so a missing approval is never hidden for longer than the run.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • answered: Waiting on checks counts every running check, not only required ones

Correct, and deliberate: per-check required-ness is not in the payload the panel reads (bucket and status only), so counting only required checks would need a new provider read this PR does not add.
The detail does not promise the checks are the blocker; it reads Merging unblocks once required checks pass and any required reviews are approved., and with nothing running the banner reverts to the generic branch-protection copy.

@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Oct 16, 2026
@kyleseaman
kyleseaman force-pushed the feat/pr-checks-running-clarity branch from d227c9a to dae1ffc Compare October 16, 2026 20:36
@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: checkInFlight's MANUAL clause duplicates bucket === 'pending' (span=c1febe68835b)

Same class as the first-principles answer on 66e69e2 (comment 6255241326) and on da75d6b (6228056585). The clause feeds only this PR's new N running · M failed count, which renders only when running > 0 && failed > 0. A GitLab MR with one failed job and one manual gate has running = 0, so the tab keeps main's Checks running complete/total label and main's row spinner, exactly as on main: the clause never changes a state main already renders. Deleting it would make that MR's tab say 1 running · 1 failed while it waits on a person. The six other bucket === 'pending' sites are main's code, answering "is anything unfinished", which a manual gate is; this PR leaves them unchanged.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: the Checks tab count flips meaning between states (span=1f6062912951)

Ruled by the PR author (same class as spans 0ff431e59e12, 6261619ccaf9, 7adf419714f4, f64c74410111, 7fdc943c7a53 and f81d85b266c0; see comment 6255241370): main's complete/total fraction stays in every state main already rendered, and the explicit 1 running · 1 failed counts appear only while a check has failed and another still runs, the one state where the fraction is ambiguous.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: red tab label beside the amber spinner (span=2ff0c2977f6c)

Same class as span 2563111d867d (comment 6227278106): the pairing is main's, unchanged by this PR. Main's Checks tab already renders the amber pending spinner whenever a check is pending and the danger tone whenever one has failed (tone: checkCounts.failed ? 'text-danger' : ...), so a failed-while-running PR on main shows the same red label beside an amber spinner. This PR changes only the count text in that state; recolouring the tab's glyph logic is outside its goal.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: the chip's × and dot are a guess without words (span=9332bb2b0e11)

Same class as spans 683b65623862, 0f4b6446fc14, fea50c1dcbdb and 03d5281edad1 (comment 6255241297): the pair carries the tooltip and accessible name Checks failed; others still running, and the × and the dot are the chip's existing failed and pending glyphs; visible text would widen every session row, outside this PR's goal.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: make both banner headings say "Merge blocked" (span=ea5e1a5f7ab2)

Same class as spans ba513b5aace9 and 3c7052d001c5 (comment 6255241329): the amber title is main's own Merge blocked., and the red detail continues its counts as a sentence, "Merging stays blocked until required checks pass and any required reviews are approved."; both say blocked and name the same two conditions. They are two different states (a failure vs. nothing failed yet), so the red one leads with the failure, and rewording would re-translate 12 catalogs for no change in meaning.

@kyleseaman
kyleseaman force-pushed the feat/pr-checks-running-clarity branch from cb27ad6 to edb872d Compare October 19, 2026 17:27
@kyleseaman

Copy link
Copy Markdown
Contributor Author

Rebased onto main 3e39e7a (head edb872d) to pick up #22178, which clears main's Harness Parity Gate red (the proven claude MEMBER_CAPABILITIES allowlist entry) that failed this PR's previous round. No conflicts; range-diff vs cb27ad6 shows the patch identical. Scoped backend tests (1038), flake8/black/isort/mypy, check_harness_parity.py, tsc, eslint, focused vitest, i18n style and i18n:check pass locally.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: the chip's failed-and-running state has only a hover title (span=6bcb10713419)

Same class as spans 9332bb2b0e11, 683b65623862, 0f4b6446fc14, fea50c1dcbdb and 03d5281edad1 (comments 6271837616, 6255241297): the pair carries the tooltip and accessible name Checks failed; others still running, and the × and the dot are the chip's existing failed and pending glyphs; visible text would widen every session row, outside this PR's goal.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: render the nothing-failed Checks tab as N running · M failed too (span=44a18e6fb6b2)

Ruled by the PR author (same class as spans 1f6062912951, 0ff431e59e12, 6261619ccaf9, 7adf419714f4, f64c74410111, 7fdc943c7a53 and f81d85b266c0; comments 6271836793, 6255241370): main's complete/total fraction stays in every state main already rendered, and the explicit 1 running · 1 failed counts appear only while a check has failed and another still runs, the one state where the fraction is ambiguous.

@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge and removed readiness: checking Automated validation is still running labels Oct 19, 2026
@kyleseaman
kyleseaman force-pushed the feat/pr-checks-running-clarity branch from edb872d to d34a549 Compare October 20, 2026 21:35
@kyleseaman

Copy link
Copy Markdown
Contributor Author

Rebased onto main (bdb1c87); new head d34a549. One conflict: docs/system-specs/modules/learn-cron-dashboard.md. Main added the handlers/core.py bullet right after the handlers/source_providers.py bullet this PR edits (ciPending in the chip-status shape). Kept both: main's new bullet plus this PR's unchanged edit. Range-diff shows the PR's own patch unchanged; tsc, eslint, i18n:check, scoped backend tests and the PR's frontend tests pass locally. Prior head kept at backup/feat/pr-checks-running-clarity-edb872dbd4.

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: checking Automated validation is still running labels Oct 20, 2026
@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: the Checks tab mixes a running count with a done count (span=4818a12824fb)

Ruled by the PR author (same class as spans 44a18e6fb6b2, 1f6062912951, 0ff431e59e12, 6261619ccaf9, 7adf419714f4, f64c74410111, 7fdc943c7a53 and f81d85b266c0; comments 6273632693, 6271836793, 6255241370): main's complete/total fraction stays in every state main already rendered, and the explicit 1 running · 1 failed counts appear only while a check has failed and another still runs, the one state where the fraction is ambiguous. Unchanged by this rebase.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: the chip's × plus dot is only guessed (span=2ac677c0cc3f)

Same class as spans 6bcb10713419, 9332bb2b0e11, 683b65623862, 0f4b6446fc14, fea50c1dcbdb and 03d5281edad1 (comments 6273632370, 6271837616, 6255241297): the pair carries the tooltip and accessible name Checks failed; others still running, and the × and the dot are the chip's existing failed and pending glyphs; visible text would widen every session row, outside this PR's goal.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • rebutted: start the failed-plus-running detail with the passed count (span=f81d6f62d647)

Same class as span 8a7d9bca9fdc (comment 6226585378): the banner names what blocks the merge, the failed and the still-running checks out of the total; the passed count is the tab's job and is visible beside it. Adding it would make the banner restate the tab and grow the string in every catalog for no new fact.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • answered: ciPending counts every pending bucket while the panel's checkInFlight leaves out MANUAL

Same answer as comments 6226585546 and 6228056585, still true at this head: they cannot disagree on the providers that get the marker. GitHub check rows carry the GraphQL status vocabulary and Bitbucket Server rows carry build states; neither ever reports MANUAL, which only gitlab.py _gitlab_check emits. GitLab never gets ciPending (CI_PENDING_PROVIDERS in projection.py is GitHub and Bitbucket Server), so the panel's MANUAL rule only shapes GitLab's banner and tab count, where the chip has no marker to match.

@kyleseaman

Copy link
Copy Markdown
Contributor Author
  • answered: GitHub blocked with a check running replaces the branch-protection copy even when a missing approval blocks

Same answer as comments 6228576992 and 6227278244, still true at this head: the title stays main's Merge blocked., and the detail names both gates: "Checks are still running. Merging unblocks once required checks pass and any required reviews are approved.", so a missing review is stated, not hidden. Once nothing is in flight, the banner returns to main's branch-protection copy unchanged.

@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Oct 20, 2026
@kyleseaman
kyleseaman force-pushed the feat/pr-checks-running-clarity branch from d34a549 to f5ac587 Compare October 21, 2026 14:03
@kyleseaman

Copy link
Copy Markdown
Contributor Author

Rebased onto main 8241440 (head d34a549 -> f5ac587). One conflict: docs/system-specs/modules/learn-cron-dashboard.md, where main rewrote the adjacent Chat Folders paragraph next to this PR's ciPending edits to the Pull-request sources paragraph; kept main's text and re-applied the ciPending additions. The PR's own changed lines are byte-identical to d34a549. Backup: backup/feat/pr-checks-running-clarity-d34a549a36. Local gates: tsc, eslint, i18n:check, flake8, mypy, focused vitest (129) and pytest (139), push_guard all pass.

@iamwhatever

Copy link
Copy Markdown
Contributor

HOLD at f5ac587.

The change still reads sound after the rebase. _record_ci is the one writer of ci + ciPending, plugin reads drop the marker, and a degraded read carries both together. Opus and GPT report no findings, and CI is green.

To clear:

  1. Maintainer sign-off on the app-token scope edit. ws_event_scope.py#L1423 adds ciPending to the fields stripped from app-token frames, in step with source_links.py#L238. App tokens still get no provider status, but token-scope edits go to a human.
  2. Dispositions for this head's lane notes:
    • Design: a degraded read can keep "others still running" after the checks finish (chip_refresh.py#L339).
    • UX: the shrunk amber dot reads green in light theme.
    • UX: lead the red title with the counts.
    • UX: the two Checks-tab count forms. The earlier class rebuttal can be re-posted for this head.

— crew-pr-reviewer (r117)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants