fix(pull-requests): reuse github api reads by maria-rcks · Pull Request #9176 · pingdotgg/t3code

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

maria-rcks · 2026-09-02T05:58:57Z

Linked pull request polling and default pull request pages fan out into full-detail reads, repeated viewer lookups, empty-search fallbacks, and unbounded gh subprocess bursts; production trace sampling tied most observed failures to host rate pauses and gh spawn EAGAIN. This adds a cached summary RPC, single-flight viewer and search reuse, bounded stale-on-transient reads, repository-access caching, deferred diffs, and global gh and VCS concurrency caps. Synthetic normal-workflow measurements now take 5 host reads instead of 11 for a cold default page, 1 instead of 4 for linked PR status, 0 repeat reads inside cache windows, and cap a 32-request gh burst at 4 concurrent processes; 225 focused tests and all five affected package typechecks pass. Implemented with gpt-5.6-sol through the T3 Code Codex harness.

Note

Medium Risk
Changes pull-request read paths, cache/invalidation semantics, and global gh concurrency, which can alter freshness and latency under load or host errors.

Overview
Introduces a PullRequestSummary shape and pullRequests.summary RPC so linked threads and settlement logic can refresh PR state without full detail reads. Mobile, web, and client-runtime linked-PR atoms now call summary on a 60s refresh cadence instead of detail.

On the server, PullRequestService.summary caches successful reads, optionally serves last-good data for up to 10 minutes after transient provider failures (unless recoverTransientFailure: false), and GitHub uses a narrow getPullRequestSummary path via gh pr view. Listing work is cheaper through single-flight viewer lookups, a search-visibility map that skips redundant per-repo fallbacks, and cached repository access on detail. VcsProcess caps concurrent VCS/gh subprocesses and the shared limiter is wired at the app layer.

The PR list UI waits for the main list before firing authored/reviewing partition reads, and the detail panel drops automatic diff warm-up on open.

^{Reviewed by Cursor Bugbot for commit 7dc9917. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add `PullRequestSummary` contract and reuse GitHub API reads through caching and narrow lookups

Adds a compact PullRequestSummary schema and a narrow getChangeRequestSummary provider operation so linked PR reads fetch fewer fields than full detail
Adds server-side caching: concurrent viewer lookups coalesce, host-search results mark repositories visible for ten minutes, and failed summary/detail reads serve last-good values for up to ten minutes
Migrates ThreadSettlementReactor linked-PR lookups from PullRequestService.detail to .summary with transient-failure recovery disabled
Limits VCS process concurrency to 8 general permits and 4 GitHub (gh) permits per VcsProcess instance; moves the shared VCS process layer to the server application boundary
Migrates client linked-PR state to the pullRequestsSummary RPC with a 60-second stale/refresh interval (previously 15s/30s)
Risk: PullRequestSummary omits fields present in the detail shape; any consumer expecting full detail from the summary RPC must fall back to the detail RPC. VcsProcess.run now caps gh commands at 4 concurrent, which may queue bursts that previously ran unbounded

^{Macroscope summarized 7dc9917.}

macroscopeapp · 2026-09-02T06:09:34Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a substantial cross-cutting change that alters pull-request freshness, caching, fallback behavior, process scheduling, and linked-thread polling defaults. It also adds an authenticated RPC and modifies authorization code, so the runtime and access-surface changes require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc66aab. Configure here.}

Renumbers the v2 migrations 044-052 to 045-053 behind main's 044_ClearAutomaticProjectModelDefaults, and adopts main's sticky new-thread selection (#9164), local-only worktree bases in the v2 launch path (#8751), the PR summary read for settlement (#9176), Claude per-cwd skills (#9210), the provider editor redesign with the branch's dedicated environment fields re-grafted (#8508), and the client half of continue-threads-across-restart (#9167). The server-side continuation markers stay unported: they live in the v1 session directory, and v2 recovery terminalizes running runs on restart, so the capability is withheld until the v2 runtime carries them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…continuation, provider editor redesign, context meter opt-in, linked-PR summaries) Brings the fork up to origin/main e0da6c6..b57726c. Highlights: continue active threads across server self-updates (pingdotgg#9167), provider editor and models list redesign (pingdotgg#8508), opt-in context window meter (pingdotgg#9190), linked-PR summary RPC with last-good caching (pingdotgg#9176), relay credential refresh before expiry (pingdotgg#9178), composer draft preserved through worktree setup (pingdotgg#9197), project skill discovery for Claude (pingdotgg#9210), copy-path on diff headers (pingdotgg#2403), mobile long-press file references (pingdotgg#9258). Conflict resolutions (14 files), keep-both unless noted: - RpcAuthorization / contracts rpc / ws.ts: fork's PR rank, cherry-pick and upstream-release RPCs sit beside upstream's new pullRequestsSummary. - ws.ts route layer: fork's computer task/view brokers plus upstream's serverSelfUpdate wrapped in withRunningThreadContinuation. - PullRequestService: fork's upstream-project reads and scoped diff epochs kept; upstream's summary read, refCacheKey and viewer-flight invalidation added. invalidate() now scopes diff-only invalidation AND clears viewer flights on a whole-workspace refresh. Test file keeps both new suites. - client-runtime pullRequests: fork's PR stack atoms kept, TTL constant takes upstream's new name. - ChatMarkdown: upstream's one-character gutter rule replaces the fork's equivalent digitWidth fix. - ChatComposer: fork's voice-session button kept; context meter now honours settings.contextWindowMeterEnabled. - ConnectionsSettings: fork's EnvironmentLabelControl kept with upstream's min-w-0 truncate. - mobile ThreadFeed: upstream's markdownLinkHandlers plus the fork's onCancelQueuedMessage. - ClaudeDriver imports: fork's superset. build-desktop-artifact: distro and arch both threaded through. Integration: upstream's restart continuation and the fork's resume-on-restart both ran at boot on the same orphans. The continuation-marker helpers move to provider/serverUpdateContinuation.ts and SessionStartupReconciler now defers any binding carrying the marker to upstream's pass (logged as deferred), so a thread is never continued twice. Test added. Dependencies reinstalled for the expo-sharing pin; the stale pnpm copy of @t3tools/mobile-markdown-text was what broke the mobile typecheck. Typecheck clean: contracts, shared, client-runtime, server, web, mobile (desktop exits 1 on a pre-existing warning-level diagnostic). Startup and self-update suites 36 passing; web sidebar/composer/settings suites 419 passing; fork guard script OK. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

github-actionsBot added vouch:trusted PR author is trusted by repo permissions or the VOUCHED list. size:L 100-499 changed lines (additions + deletions). labels Sep 2, 2026

macroscopeappBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/pullRequest/PullRequestService.ts Outdated
Comment threadapps/server/src/pullRequest/GitHubPullRequestCli.ts Outdated

macroscopeappBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/pullRequest/PullRequestService.ts Outdated
Comment threadapps/server/src/pullRequest/PullRequestService.ts Outdated
Comment threadapps/server/src/pullRequest/PullRequestService.ts
Comment threadapps/server/src/pullRequest/PullRequestService.ts

cursorBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/pullRequest/PullRequestService.ts Outdated

fix(pull-requests): reuse github api reads
6879910

maria-rcks force-pushed the t3code/benchmark-api-rate-limits branch from 50f36df to f183b57Compare September 2, 2026 06:12

github-actionsBot added size:XL 500-999 changed lines (additions + deletions). and removed size:L 100-499 changed lines (additions + deletions). labels Sep 2, 2026

fix(pull-requests): handle cache invalidation edges
b2b3378

maria-rcks force-pushed the t3code/benchmark-api-rate-limits branch from f183b57 to b2b3378Compare September 2, 2026 06:15

github-actionsBot added size:L 100-499 changed lines (additions + deletions). and removed size:XL 500-999 changed lines (additions + deletions). labels Sep 2, 2026

macroscopeappBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/pullRequest/PullRequestService.ts

macroscopeappBot reviewed Sep 2, 2026
View reviewed changes

Comment threadpackages/client-runtime/src/state/pullRequests.ts

macroscopeappBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/vcs/VcsProcess.ts

cursorBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/orchestration/ThreadSettlementReactor.ts

maria-rcks added 2 commits September 2, 2026 06:28

style(pull-requests): match pinned formatter
1b2312f

fix(pull-requests): close review edge cases
fc66aab

macroscopeappBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/vcs/VcsProcess.ts Outdated

cursorBot reviewed Sep 2, 2026
View reviewed changes

Comment threadapps/server/src/pullRequest/PullRequestService.ts Outdated

maria-rcks added 2 commits September 2, 2026 06:41

fix(server): share vcs process limiter
caa4bdd

fix(pull-requests): retain strict summary reads
7dc9917

maria-rcks merged commit dd6879f into pingdotgg:mainSep 2, 2026
27 checks passed

patroza mentioned this pull request Sep 2, 2026
merge: sync upstream through 70cd258d8 (8 commits) patroza/t3code#442
Merged

berghtho mentioned this pull request Sep 2, 2026
chore(upstream): sync 2026-09-02 berghtho/t3code#20
Merged