You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The initial merge stopped in 91 files. Resolved the merge while preserving upstream structure and retained fork policies. No webhook request data was used. Only the copied trusted publisher will publish this candidate; main was not pushed.
Superseded patches
Aborted the first merge and reverted these already-equivalent patches before merging again:
Removed their entries from fork/CHANGES.md. Reverts are ordinary commits; existing fork history was retained. The remaining conflicts were resolved in merge 4051939c72; follow-up adaptations are in 49114600a1e31c683c32e7c73d99678c57ea58b5.
Resolutions
Desktop and documentation: moved tray initialization into upstream's new bootstrap phase; retained hidden tray state, delayed initial maximize and update-bound flush while adopting upstream minimized-window persistence and SSH tunnel cleanup. Added the tray setting to upstream's per-key tolerant decoder. Kept both navigation-policy and preview-guest tests. Windows background-service guidance now follows upstream support and retains the separate tray instructions.
Clients and shared runtime: combined startup steering, account confirmation, context notes, one-shot compaction and Always auto-settle with upstream composer capability/workspace APIs. Scheduled tasks retain model-family following while using upstream timezone and fast-mode helpers. Timelines adopt upstream partial-answer/steer segmentation and keep completed confirmed answers visible. Existing mobile and composer fixtures now represent the expanded upstream APIs. Retained both shared exports and issue-watch activity behavior.
MCP/contracts/evals: combined fork context notes, account/stall reporting, short delegation waits and settled-only result policy with upstream newest-first budgeted timeline pages, title outcomes, background-task metadata and forwarding wrappers. Explicit forward opening reads replace old history with context notes; latest-page reads preserve recent content. Upstream AssistantMessagePhase is the schema source. Both eval toolkits and both groups of grading cases remain.
Orchestration: retained native-boundary offers/receipts, pending intent/priority, context-policy compaction, issue watches and account selection while adopting upstream stale-question guards, goal/restart scheduling, early workspace preparation and sparse projections. Restored the deferred startup promotion branch; implicit new startup steers use the running model, while persisted redeliveries retain their saved selection. User Auto after Stop keeps upstream fallback; agents retain the fork's Stop guard. Follow-up waiting tests explicitly request the fork's settled policy; every-turn remains the documented default. Added a guarded fork expression index so one-shot compaction does not scan thread history.
Claude adapter: combined fork native hook delivery, cancellation notes, transient-error recovery, quota reset retention and keep-warm state with upstream process-exit reporting, descendant capture, live model/effort updates, streaming blocks and startup preparation. Phase registration now covers the shared streaming/snapshot publisher. Retained attachment and native-fork suites. The generic failed-turn wakeup fixture uses permanent HTTP400 because fork HTTP529 errors retry.
Codex adapter: retained serialized native acceptance and day-bounded capacity/transport retries alongside upstream retry-stall watchdog, configured reasoning summaries, access-program selection, goal wakeups and compaction context restoration. Configuration reads are inside the acceptance lane. Native partial answers remain partial; streaming final phases follow upstream, but a cleanup flush cannot confirm an unfinished final answer. Updated strict phase expectations accordingly. The stalled429 regression verifies native interruption followed by a reset-less transport retry, matching the fork's explicit throttling policy. Startup tests await the request milestone before advancing their TestClock; retry wire fixtures use upstream's auto reasoning-summary default.
Provider support: ACP session reads keep their provider instance and upstream deferred-tool options. Usage windows retain both credits and zero-spend-cap metadata. OpenCode keeps fork version2/legacy records and upstream wrapper-command fallback, including records with no serve arguments. Codex native usage reads follow upstream's confirmed-route classifier while retaining fork probe regressions. Dev runner keeps live-home protection and upstream memory/bind handling. Fetch and index-refresh timeouts both remain.
SQLite ownership: final review found that merely appending fork DDL after upstream migration setup bypassed its CLI ownership scope. Reproduced this with existing ownership tests extended for missing fork indexes/table. Upstream and fork DDL now share the ownership-protected migration effect; a fully-current CLI read performs no fork DDL or owner acquisition. No upstream migration IDs were allocated.
Test merges distorted by upstream removing redundant Effect.scoped wrappers were aligned through temporary AST/scope normalization, retaining the actual test behaviors. Temporary merge/debug aids are outside the checkout; no debug code is committed.
Critical Claude streaming/phase/steer/wakeup selection: 13 passed; the separately isolated permanent-error wakeup case also passed. Full518-case Claude suite was killed with exit137 under memory pressure; it is skipped locally for that environment reason, not counted as passed.
Web/mobile/shared-client focused cases: 426/426 passed; web, mobile and client-runtime typechecks and scoped client lint passed.
Contracts, desktop, provider-acp, provider-core and provider-opencode package typechecks passed.
MCP + provider-switch coverage: 137 cases passed across serial and isolated runs. Two contention-related timeouts passed isolated without changes. Contracts6, activity8, scheduler23 and the eval runner passed.
Eight orchestration suites: 325 initially passed; the two failures were fixed and passed in the final16-case regression selection. Seven full suites were green (projection61, delegated completion65, idle compaction39, settlement44, positions3, Claude automatic delivery6, HTTP2); other steering cases passed in the full run.
Provider probing/usage, OpenCode ledger and dev runner: 83 passed. Agent context budget: 2 passed, with committed measured values.
Desktop/persistence initial focused run: 66 passed. Final SQLite ownership suite after the additional fix: 8/8 passed, with a failing regression first.
Server typecheck: passed exit0 after resource pressure subsided (serial run; existing schema-brand suggestions only).
corepack pnpm exec vp fmt --check: passed on all7380 matched files. Final diff/state/ancestry checks: clean checkout, no unresolved entries or merge/revert state, both requested parents are ancestors; diff whitespace findings verified upstream-only as detailed below.
Earlier parallel checks coincided with the container's12GiB memory limit and host load, producing OOM exits and worker/time-out failures. Superseded owned runs were stopped; later tests used maxWorkers1. No timeouts, assertions or type checks were weakened. Actual semantic failures were reproduced and fixed; environment deaths are identified separately above.
git diff --check reports whitespace only inside upstream-owned patch files patches/@cursor__sdk@1.2.0.patch and patches/@pierre__diffs@1.7.0-beta.3.patch. Their candidate blobs match the requested upstream blobs exactly (cf657be869d5c688ad99443f02097f1ac3d8559c, 5704f9c799cf0dbd8ceb2486da1e8e49fe60f49c); left unchanged per procedure, not counted as failed validation.
Review focus
Inspect the boundary between fork native steering and upstream deferred startup/Stop; every-turn versus settled child delivery; context-note versus newest-first timeline reads; streamed versus confirmed answer phases; and CLI ownership of fork schema additions. Upstream fixes native usage routing more broadly than the retained unsupported-read probe patch: please check whether that fork patch can now be fully retired. OpenCode's legacy/version2 record compatibility and wrapper fallback also deserve review.
The only resolver-authored change under fork/ is fork/CHANGES.md (two superseded entries removed and formatting). No resolver-authored workflow/action change; neither the procedure nor fork/scripts/t3-resolver/ was edited. All files in the conflict inventory follow.
The initial merge stopped in 91 files. Resolved the merge while preserving upstream structure and retained fork policies. No webhook request data was used. Only the copied trusted publisher will publish this candidate; main was not pushed.
Superseded patches
Aborted the first merge and reverted these already-equivalent patches before merging again:
Removed their entries from fork/CHANGES.md. Reverts are ordinary commits; existing fork history was retained. The remaining conflicts were resolved in merge 4051939c72; follow-up adaptations are in 49114600a1e31c683c32e7c73d99678c57ea58b5.
Resolutions
Desktop and documentation: moved tray initialization into upstream's new bootstrap phase; retained hidden tray state, delayed initial maximize and update-bound flush while adopting upstream minimized-window persistence and SSH tunnel cleanup. Added the tray setting to upstream's per-key tolerant decoder. Kept both navigation-policy and preview-guest tests. Windows background-service guidance now follows upstream support and retains the separate tray instructions.
Clients and shared runtime: combined startup steering, account confirmation, context notes, one-shot compaction and Always auto-settle with upstream composer capability/workspace APIs. Scheduled tasks retain model-family following while using upstream timezone and fast-mode helpers. Timelines adopt upstream partial-answer/steer segmentation and keep completed confirmed answers visible. Existing mobile and composer fixtures now represent the expanded upstream APIs. Retained both shared exports and issue-watch activity behavior.
MCP/contracts/evals: combined fork context notes, account/stall reporting, short delegation waits and settled-only result policy with upstream newest-first budgeted timeline pages, title outcomes, background-task metadata and forwarding wrappers. Explicit forward opening reads replace old history with context notes; latest-page reads preserve recent content. Upstream AssistantMessagePhase is the schema source. Both eval toolkits and both groups of grading cases remain.
Orchestration: retained native-boundary offers/receipts, pending intent/priority, context-policy compaction, issue watches and account selection while adopting upstream stale-question guards, goal/restart scheduling, early workspace preparation and sparse projections. Restored the deferred startup promotion branch; implicit new startup steers use the running model, while persisted redeliveries retain their saved selection. User Auto after Stop keeps upstream fallback; agents retain the fork's Stop guard. Follow-up waiting tests explicitly request the fork's settled policy; every-turn remains the documented default. Added a guarded fork expression index so one-shot compaction does not scan thread history.
Claude adapter: combined fork native hook delivery, cancellation notes, transient-error recovery, quota reset retention and keep-warm state with upstream process-exit reporting, descendant capture, live model/effort updates, streaming blocks and startup preparation. Phase registration now covers the shared streaming/snapshot publisher. Retained attachment and native-fork suites. The generic failed-turn wakeup fixture uses permanent HTTP400 because fork HTTP529 errors retry.
Codex adapter: retained serialized native acceptance and day-bounded capacity/transport retries alongside upstream retry-stall watchdog, configured reasoning summaries, access-program selection, goal wakeups and compaction context restoration. Configuration reads are inside the acceptance lane. Native partial answers remain partial; streaming final phases follow upstream, but a cleanup flush cannot confirm an unfinished final answer. Updated strict phase expectations accordingly. The stalled429 regression verifies native interruption followed by a reset-less transport retry, matching the fork's explicit throttling policy. Startup tests await the request milestone before advancing their TestClock; retry wire fixtures use upstream's auto reasoning-summary default.
Provider support: ACP session reads keep their provider instance and upstream deferred-tool options. Usage windows retain both credits and zero-spend-cap metadata. OpenCode keeps fork version2/legacy records and upstream wrapper-command fallback, including records with no serve arguments. Codex native usage reads follow upstream's confirmed-route classifier while retaining fork probe regressions. Dev runner keeps live-home protection and upstream memory/bind handling. Fetch and index-refresh timeouts both remain.
SQLite ownership: final review found that merely appending fork DDL after upstream migration setup bypassed its CLI ownership scope. Reproduced this with existing ownership tests extended for missing fork indexes/table. Upstream and fork DDL now share the ownership-protected migration effect; a fully-current CLI read performs no fork DDL or owner acquisition. No upstream migration IDs were allocated.
Test merges distorted by upstream removing redundant Effect.scoped wrappers were aligned through temporary AST/scope normalization, retaining the actual test behaviors. Temporary merge/debug aids are outside the checkout; no debug code is committed.
Critical Claude streaming/phase/steer/wakeup selection: 13 passed; the separately isolated permanent-error wakeup case also passed. Full518-case Claude suite was killed with exit137 under memory pressure; it is skipped locally for that environment reason, not counted as passed.
Web/mobile/shared-client focused cases: 426/426 passed; web, mobile and client-runtime typechecks and scoped client lint passed.
Contracts, desktop, provider-acp, provider-core and provider-opencode package typechecks passed.
MCP + provider-switch coverage: 137 cases passed across serial and isolated runs. Two contention-related timeouts passed isolated without changes. Contracts6, activity8, scheduler23 and the eval runner passed.
Eight orchestration suites: 325 initially passed; the two failures were fixed and passed in the final16-case regression selection. Seven full suites were green (projection61, delegated completion65, idle compaction39, settlement44, positions3, Claude automatic delivery6, HTTP2); other steering cases passed in the full run.
Provider probing/usage, OpenCode ledger and dev runner: 83 passed. Agent context budget: 2 passed, with committed measured values.
Desktop/persistence initial focused run: 66 passed. Final SQLite ownership suite after the additional fix: 8/8 passed, with a failing regression first.
Server typecheck: passed exit0 after resource pressure subsided (serial run; existing schema-brand suggestions only).
corepack pnpm exec vp fmt --check: passed on all7380 matched files. Final diff/state/ancestry checks: clean checkout, no unresolved entries or merge/revert state, both requested parents are ancestors; diff whitespace findings verified upstream-only as detailed below.
Earlier parallel checks coincided with the container's12GiB memory limit and host load, producing OOM exits and worker/time-out failures. Superseded owned runs were stopped; later tests used maxWorkers1. No timeouts, assertions or type checks were weakened. Actual semantic failures were reproduced and fixed; environment deaths are identified separately above.
git diff --check reports whitespace only inside upstream-owned patch files patches/@cursor__sdk@1.2.0.patch and patches/@pierre__diffs@1.7.0-beta.3.patch. Their candidate blobs match the requested upstream blobs exactly (cf657be869d5c688ad99443f02097f1ac3d8559c, 5704f9c799cf0dbd8ceb2486da1e8e49fe60f49c); left unchanged per procedure, not counted as failed validation.
Review focus
Inspect the boundary between fork native steering and upstream deferred startup/Stop; every-turn versus settled child delivery; context-note versus newest-first timeline reads; streamed versus confirmed answer phases; and CLI ownership of fork schema additions. Upstream fixes native usage routing more broadly than the retained unsupported-read probe patch: please check whether that fork patch can now be fully retired. OpenCode's legacy/version2 record compatibility and wrapper fallback also deserve review.
The only resolver-authored change under fork/ is fork/CHANGES.md (two superseded entries removed and formatting). No resolver-authored workflow/action change; neither the procedure nor fork/scripts/t3-resolver/ was edited. All files in the conflict inventory follow.
Hosted validation fix round 1
The first hosted run 43584558785 failed lint and focused web/provider/server cases. These are fixed in ordinary commit c1eda99e9fa9dcc5b6c03aed7f451172b9a3ca62; the existing merge and initial adaptation commits are retained. Review PR: #92.
Restored use of upstream structured Claude process-exit telemetry: a killed process reports its signal/code without stderr and ends the turn, while ordinary transient failures retain fork retry behavior. This also fixes the unused-helper lint error. The exact signal test failed locally before the fix and passes afterward.
Cursor human input now reaches native steering directly because Cursor has no Claude boundary hook. Explicit agent requests and agent-authored queued input remain protected. The native replay failed locally before the fix and passes; the existing queue-promotion table now includes three Cursor cases and retains all original agency/boundary cases.
Startup reconciliation reads queued/running records and scoped messages/turns instead of all historical runs. Implicit startup selection retains the target only within the same provider instance, preserving cross-instance context policies.
Restored the fork's zero-text-offset acknowledgment guard alongside upstream's past-end guard, so a suffix-only read cannot acknowledge an unseen delegated result.
Held tasks in the fork can be idle. Atomic cancellation now recognizes that state and records the owned cancelled run before finalization; the integration asserts the exact run marker and disposed delivery.
Fixtures now provide real pending child questions, actual rejected receipt targets, forwarded archive callbacks, scoped Cursor MCP reads, registered foreign projects, explicit opening-read cursors and current argument-error text. Queue browser fixtures include delivery metadata and use the fork's existing message label; quote/citation assertions remain.
Upstream recovery intentionally re-offers open completion reservations. The large integration now checks exact logical reservation counts (one, two or three at each phase), nonempty reservation identities, identical payloads for every retry, persisted message/task ownership, and cancellation/approval states. It waits through the existing PubSub milestone. Raw retry attempts observed during recovery are not treated as distinct user notifications. No queue deduplication, checkpoint recovery, generation guard or original lifecycle assertion was disabled. This follows the unchanged upstream recovery path and its explicit repeated-recovery regression. A separate exact count of zero still forbids any offer after disposal.
Fix-round validation, all successful:
Claude signal + Cursor steering replay: 2 focused regressions passed; scoped Cursor workspace MCP case: 1 passed.
Startup/performance/cross-instance/question cases: 7 passed across four files.
Cursor promotion/boundary/agency/Stop safety: 23 passed, including all three new Cursor rows.
Queued-run browser file: all 3 tests passed.
Full MCP end-to-end integration: passed on the clean final fixture after logical-reservation alignment (14.46s).
Server, provider-cursor and web typechecks: passed serially, with existing schema-brand suggestions only.
Changed-file lint, full vp fmt --check (7380 files), and changed-content whitespace checks: passed. Only the previously documented exact-upstream patch whitespace remains in the full base-to-candidate comparison.
All tests used --maxWorkers=1. Owned test/typecheck processes finished before publication; no temporary tool/error logging remains. The earlier source/fixture report and initial conflict inventory below remain applicable.
The initial merge stopped in 91 files. Resolved the merge while preserving upstream structure and retained fork policies. No webhook request data was used. Only the copied trusted publisher will publish this candidate; main was not pushed.
Superseded patches
Aborted the first merge and reverted these already-equivalent patches before merging again:
Removed their entries from fork/CHANGES.md. Reverts are ordinary commits; existing fork history was retained. The remaining conflicts were resolved in merge 4051939c72; follow-up adaptations are in 49114600a1e31c683c32e7c73d99678c57ea58b5.
Resolutions
Desktop and documentation: moved tray initialization into upstream's new bootstrap phase; retained hidden tray state, delayed initial maximize and update-bound flush while adopting upstream minimized-window persistence and SSH tunnel cleanup. Added the tray setting to upstream's per-key tolerant decoder. Kept both navigation-policy and preview-guest tests. Windows background-service guidance now follows upstream support and retains the separate tray instructions.
Clients and shared runtime: combined startup steering, account confirmation, context notes, one-shot compaction and Always auto-settle with upstream composer capability/workspace APIs. Scheduled tasks retain model-family following while using upstream timezone and fast-mode helpers. Timelines adopt upstream partial-answer/steer segmentation and keep completed confirmed answers visible. Existing mobile and composer fixtures now represent the expanded upstream APIs. Retained both shared exports and issue-watch activity behavior.
MCP/contracts/evals: combined fork context notes, account/stall reporting, short delegation waits and settled-only result policy with upstream newest-first budgeted timeline pages, title outcomes, background-task metadata and forwarding wrappers. Explicit forward opening reads replace old history with context notes; latest-page reads preserve recent content. Upstream AssistantMessagePhase is the schema source. Both eval toolkits and both groups of grading cases remain.
Orchestration: retained native-boundary offers/receipts, pending intent/priority, context-policy compaction, issue watches and account selection while adopting upstream stale-question guards, goal/restart scheduling, early workspace preparation and sparse projections. Restored the deferred startup promotion branch; implicit new startup steers use the running model, while persisted redeliveries retain their saved selection. User Auto after Stop keeps upstream fallback; agents retain the fork's Stop guard. Follow-up waiting tests explicitly request the fork's settled policy; every-turn remains the documented default. Added a guarded fork expression index so one-shot compaction does not scan thread history.
Claude adapter: combined fork native hook delivery, cancellation notes, transient-error recovery, quota reset retention and keep-warm state with upstream process-exit reporting, descendant capture, live model/effort updates, streaming blocks and startup preparation. Phase registration now covers the shared streaming/snapshot publisher. Retained attachment and native-fork suites. The generic failed-turn wakeup fixture uses permanent HTTP400 because fork HTTP529 errors retry.
Codex adapter: retained serialized native acceptance and day-bounded capacity/transport retries alongside upstream retry-stall watchdog, configured reasoning summaries, access-program selection, goal wakeups and compaction context restoration. Configuration reads are inside the acceptance lane. Native partial answers remain partial; streaming final phases follow upstream, but a cleanup flush cannot confirm an unfinished final answer. Updated strict phase expectations accordingly. The stalled429 regression verifies native interruption followed by a reset-less transport retry, matching the fork's explicit throttling policy. Startup tests await the request milestone before advancing their TestClock; retry wire fixtures use upstream's auto reasoning-summary default.
Provider support: ACP session reads keep their provider instance and upstream deferred-tool options. Usage windows retain both credits and zero-spend-cap metadata. OpenCode keeps fork version2/legacy records and upstream wrapper-command fallback, including records with no serve arguments. Codex native usage reads follow upstream's confirmed-route classifier while retaining fork probe regressions. Dev runner keeps live-home protection and upstream memory/bind handling. Fetch and index-refresh timeouts both remain.
SQLite ownership: final review found that merely appending fork DDL after upstream migration setup bypassed its CLI ownership scope. Reproduced this with existing ownership tests extended for missing fork indexes/table. Upstream and fork DDL now share the ownership-protected migration effect; a fully-current CLI read performs no fork DDL or owner acquisition. No upstream migration IDs were allocated.
Test merges distorted by upstream removing redundant Effect.scoped wrappers were aligned through temporary AST/scope normalization, retaining the actual test behaviors. Temporary merge/debug aids are outside the checkout; no debug code is committed.
Critical Claude streaming/phase/steer/wakeup selection: 13 passed; the separately isolated permanent-error wakeup case also passed. Full518-case Claude suite was killed with exit137 under memory pressure; it is skipped locally for that environment reason, not counted as passed.
Web/mobile/shared-client focused cases: 426/426 passed; web, mobile and client-runtime typechecks and scoped client lint passed.
Contracts, desktop, provider-acp, provider-core and provider-opencode package typechecks passed.
MCP + provider-switch coverage: 137 cases passed across serial and isolated runs. Two contention-related timeouts passed isolated without changes. Contracts6, activity8, scheduler23 and the eval runner passed.
Eight orchestration suites: 325 initially passed; the two failures were fixed and passed in the final16-case regression selection. Seven full suites were green (projection61, delegated completion65, idle compaction39, settlement44, positions3, Claude automatic delivery6, HTTP2); other steering cases passed in the full run.
Provider probing/usage, OpenCode ledger and dev runner: 83 passed. Agent context budget: 2 passed, with committed measured values.
Desktop/persistence initial focused run: 66 passed. Final SQLite ownership suite after the additional fix: 8/8 passed, with a failing regression first.
Server typecheck: passed exit0 after resource pressure subsided (serial run; existing schema-brand suggestions only).
corepack pnpm exec vp fmt --check: passed on all7380 matched files. Final diff/state/ancestry checks: clean checkout, no unresolved entries or merge/revert state, both requested parents are ancestors; diff whitespace findings verified upstream-only as detailed below.
Earlier parallel checks coincided with the container's12GiB memory limit and host load, producing OOM exits and worker/time-out failures. Superseded owned runs were stopped; later tests used maxWorkers1. No timeouts, assertions or type checks were weakened. Actual semantic failures were reproduced and fixed; environment deaths are identified separately above.
git diff --check reports whitespace only inside upstream-owned patch files patches/@cursor__sdk@1.2.0.patch and patches/@pierre__diffs@1.7.0-beta.3.patch. Their candidate blobs match the requested upstream blobs exactly (cf657be869d5c688ad99443f02097f1ac3d8559c, 5704f9c799cf0dbd8ceb2486da1e8e49fe60f49c); left unchanged per procedure, not counted as failed validation.
Review focus
Inspect the boundary between fork native steering and upstream deferred startup/Stop; every-turn versus settled child delivery; context-note versus newest-first timeline reads; streamed versus confirmed answer phases; and CLI ownership of fork schema additions. Upstream fixes native usage routing more broadly than the retained unsupported-read probe patch: please check whether that fork patch can now be fully retired. OpenCode's legacy/version2 record compatibility and wrapper fallback also deserve review.
The only resolver-authored change under fork/ is fork/CHANGES.md (two superseded entries removed and formatting). No resolver-authored workflow/action change; neither the procedure nor fork/scripts/t3-resolver/ was edited. All files in the conflict inventory follow.
Hosted validation fix round 1
The first hosted run 43584558785 failed lint and focused web/provider/server cases. These are fixed in ordinary commit c1eda99e9fa9dcc5b6c03aed7f451172b9a3ca62; the existing merge and initial adaptation commits are retained. Review PR: #92.
Restored use of upstream structured Claude process-exit telemetry: a killed process reports its signal/code without stderr and ends the turn, while ordinary transient failures retain fork retry behavior. This also fixes the unused-helper lint error. The exact signal test failed locally before the fix and passes afterward.
Cursor human input now reaches native steering directly because Cursor has no Claude boundary hook. Explicit agent requests and agent-authored queued input remain protected. The native replay failed locally before the fix and passes; the existing queue-promotion table now includes three Cursor cases and retains all original agency/boundary cases.
Startup reconciliation reads queued/running records and scoped messages/turns instead of all historical runs. Implicit startup selection retains the target only within the same provider instance, preserving cross-instance context policies.
An initial zero-text-offset restoration addressed a stale fork fixture but was superseded in fix round2 after history confirmed upstream intentionally supports paginated result acknowledgment; the final behavior is detailed below.
Held tasks in the fork can be idle. Atomic cancellation now recognizes that state and records the owned cancelled run before finalization; the integration asserts the exact run marker and disposed delivery.
Fixtures now provide real pending child questions, actual rejected receipt targets, forwarded archive callbacks, scoped Cursor MCP reads, registered foreign projects, explicit opening-read cursors and current argument-error text. Queue browser fixtures include delivery metadata and use the fork's existing message label; quote/citation assertions remain.
Upstream recovery intentionally re-offers open completion reservations. The large integration now checks exact logical reservation counts (one, two or three at each phase), nonempty reservation identities, identical payloads for every retry, persisted message/task ownership, and cancellation/approval states. It waits through the existing PubSub milestone. Raw retry attempts observed during recovery are not treated as distinct user notifications. No queue deduplication, checkpoint recovery, generation guard or original lifecycle assertion was disabled. This follows the unchanged upstream recovery path and its explicit repeated-recovery regression. A separate exact count of zero still forbids any offer after disposal.
Fix-round validation, all successful:
Claude signal + Cursor steering replay: 2 focused regressions passed; scoped Cursor workspace MCP case: 1 passed.
Startup/performance/cross-instance/question cases: 7 passed across four files.
Cursor promotion/boundary/agency/Stop safety: 23 passed, including all three new Cursor rows.
Queued-run browser file: all 3 tests passed.
Full MCP end-to-end integration: passed on the clean final fixture after logical-reservation alignment (14.46s).
Server, provider-cursor and web typechecks: passed serially, with existing schema-brand suggestions only.
Changed-file lint, full vp fmt --check (7380 files), and changed-content whitespace checks: passed. Only the previously documented exact-upstream patch whitespace remains in the full base-to-candidate comparison.
All tests used --maxWorkers=1. Owned test/typecheck processes finished before publication; no temporary tool/error logging remains. The earlier source/fixture report and initial conflict inventory below remain applicable.
Hosted validation fix round 2 — final candidate
The second hosted run 43589689173 reported two remaining failing tests. Both were reproduced locally before changes, and corrected in ordinary commit 3178120fcefbfb9c9436323a0da2124104b37da0 without changing the upstream regressions.
Paged result acknowledgment: history proves the zero-offset guard was old upstream behavior, not a fork policy. Upstream commit 85b160658f (fix(server): keep thread tool responses small by default pingdotgg/t3code#29268) removed it for500-character default/paginated reads; 6237824a2c (fix(server): t3_thread_read results fit MCP clients' output limit pingdotgg/t3code#33455) retains empty/past-end safety. Both commits are ancestors of the requested upstream SHA. I incorrectly restored the obsolete condition in round1 based on a fixture that had actually read the entire result as a prefix plus remainder. The final source restores upstream's nonempty completed-page rule. The fork integration now verifies the complete reconstructed result is acknowledged and adds a strict check that an empty past-end read does not acknowledge it; truncated-prefix delivery stays pending. No read-coverage state, API, or upstream-test weakening was introduced.
Runless native threads: upstream fb2e5a0128 (fix(client-runtime): imported threads that never ran no longer sit in Working as Waiting pingdotgg/t3code#33042), also present in the requested merge, intentionally removed an active-provider guard from runtime derivation for imported/runless threads. The merge had accidentally retained that obsolete guard while preserving fork issue-watch input. The final source adopts the upstream condition and retains issue-watch derivation. A new focused control proves watched threads with no app runs remain idle/observable and become null when the watch is removed; the unchanged provider-native subagent regression now passes.
Final-round validation:
Unchanged upstream long-child-result regression, full fork MCP integration, and acknowledgment retry unit: 3/3 passed after the reproduced failure.
Entire owning client runtime module: 34/34 passed, including the unchanged native-subagent regression and the new issue-watch control.
Server and client-runtime typechecks: passed serially, existing schema-brand suggestions only.
Changed-file lint, full formatting check (7380 matched files), and changed-content whitespace checks: passed.
Clean committed checkout, both requested ancestors, no unresolved index/merge/revert state, and unchanged protected procedure/helper tree verified again before publication. Full base comparison still has only the documented exact-upstream patch whitespace.
All owned test/typecheck processes have finished. This is the second and final permitted fix round; promotion still requires the trusted helper's hosted checks and exact-commit Codex review.
Candidate 3178120fcefbfb9c9436323a0da2124104b37da0 was not promoted automatically: Checks finished with failure after 2 fix attempts. It needs human review.
Maintainer follow-up for request 43580534179. Final candidate: 2476a907806269f7e97424613e4221bc98387116.
The upstream merge, adaptations, and two resolver CI fix rounds remain intact; their detailed summary is #91 (comment) and the preceding resolution comments. The original candidate 3178120fcefbfb9c9436323a0da2124104b37da0 passed all hosted jobs in run 43591599431 attempt 2 after retrying only the failed Web job. The initial browser connection loss remains unexplained; its named upstream-identical test also passed locally.
One independent Opus review covered the high-risk resolutions and follow-up commits. The confirmed test gap is corrected in the final commit: the integration now checks that the persisted reservation matches the offered generation and message ID and contains both completed and interrupted follow-up task IDs. This adds coverage without changing production behavior or removing assertions.
The proposed startup waiting-run regression was dismissed after direct reproduction. Widening the read still reaches an explicit rejection in dispatchSteerIntoRun for any target status other than running. Both fork base 51d334f and pinned upstream db2cc20 have that guard, so the claimed formerly successful waiting-run steering did not occur. Experimental source/test diagnostics were fully removed; the only maintainer source diff is the nine-line coverage assertion.
Final local validation: check:changed against 3178120 passed format, lint, both dependency checks, GraphQL generation check, three package typechecks, and three tests in two files, under Node 26.12.1. git diff --check and LF checks passed. The complete MCP integration passed with the new task coverage. No live provider/client/remote session was exercised.
Hosted Codex review remains quota-blocked; no credits were added or automated review metadata changed. The independent review and maintainer findings disposition are recorded at #92 (comment). Fresh hosted validation is required on this final candidate before manual promotion. Main has not been pushed by the maintainer.
Activity
Alb11747 commentedon Nov 14, 2026
Fork sync resolution — request 43580534179
Candidate:
49114600a1e31c683c32e7c73d99678c57ea58b5. Fork base:51d334f8a2107185383cf45e6250e905e44762e9. Requested upstream:db2cc2026427c5eb647a08538c12402341db0fab.The initial merge stopped in 91 files. Resolved the merge while preserving upstream structure and retained fork policies. No webhook request data was used. Only the copied trusted publisher will publish this candidate; main was not pushed.
Superseded patches
Aborted the first merge and reverted these already-equivalent patches before merging again:
4414c245461db55d9700eca3af1656f142f909b2, inherited dev-server bind settings, replaced by upstreamf9cadccadf4f6d6e37b005b7fec143c70cadbd6c(fix(scripts): dev runner ignores the parent T3 server's port and host pingdotgg/t3code#36019).c3f2943bafcdf5ad683dde3870ae4619889ba654, invalid gitfile cleanup, replaced by upstream301999c325cebee9f4330ee14ef8c3b083e67a64(fix(server): git 2.56 broken-gitfile worktrees count as non-repositories again pingdotgg/t3code#33423).Removed their entries from
fork/CHANGES.md. Reverts are ordinary commits; existing fork history was retained. The remaining conflicts were resolved in merge4051939c72; follow-up adaptations are in49114600a1e31c683c32e7c73d99678c57ea58b5.Resolutions
Test merges distorted by upstream removing redundant Effect.scoped wrappers were aligned through temporary AST/scope normalization, retaining the actual test behaviors. Temporary merge/debug aids are outside the checkout; no debug code is committed.
Validation
corepack pnpm exec vp fmt --check: passed on all7380 matched files. Final diff/state/ancestry checks: clean checkout, no unresolved entries or merge/revert state, both requested parents are ancestors; diff whitespace findings verified upstream-only as detailed below.Earlier parallel checks coincided with the container's12GiB memory limit and host load, producing OOM exits and worker/time-out failures. Superseded owned runs were stopped; later tests used maxWorkers1. No timeouts, assertions or type checks were weakened. Actual semantic failures were reproduced and fixed; environment deaths are identified separately above.
git diff --checkreports whitespace only inside upstream-owned patch filespatches/@cursor__sdk@1.2.0.patchandpatches/@pierre__diffs@1.7.0-beta.3.patch. Their candidate blobs match the requested upstream blobs exactly (cf657be869d5c688ad99443f02097f1ac3d8559c,5704f9c799cf0dbd8ceb2486da1e8e49fe60f49c); left unchanged per procedure, not counted as failed validation.Review focus
Inspect the boundary between fork native steering and upstream deferred startup/Stop; every-turn versus settled child delivery; context-note versus newest-first timeline reads; streamed versus confirmed answer phases; and CLI ownership of fork schema additions. Upstream fixes native usage routing more broadly than the retained unsupported-read probe patch: please check whether that fork patch can now be fully retired. OpenCode's legacy/version2 record compatibility and wrapper fallback also deserve review.
The only resolver-authored change under
fork/isfork/CHANGES.md(two superseded entries removed and formatting). No resolver-authored workflow/action change; neither the procedure norfork/scripts/t3-resolver/was edited. All files in the conflict inventory follow.Initial conflict inventory
apps/desktop/src/app/DesktopApp.tsapps/desktop/src/settings/DesktopAppSettings.tsapps/desktop/src/updates/DesktopUpdates.tsapps/desktop/src/window/DesktopWindow.test.tsapps/desktop/src/window/DesktopWindow.tsapps/mobile/src/features/settings/SettingsScheduledTasksRouteScreen.tsxapps/mobile/src/features/threads/ThreadSettingsSheet.tsxapps/mobile/src/lib/threadActivity.tsapps/mobile/src/state/thread-outbox-model.tsapps/mobile/src/state/use-thread-composer-state.tsapps/server/src/mcp/OrchestratorMcpService.activity.test.tsapps/server/src/mcp/OrchestratorMcpService.test.tsapps/server/src/mcp/OrchestratorMcpService.tsapps/server/src/mcp/toolkits/orchestrator/handlers.tsapps/server/src/mcp/toolkits/orchestrator/tools.tsapps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.tsapps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.tsapps/server/src/orchestration-v2/Adapters/CodexAdapterV2.test.tsapps/server/src/orchestration-v2/Adapters/CodexAdapterV2.tsapps/server/src/orchestration-v2/ClaudeAutomaticDelivery.integration.test.tsapps/server/src/orchestration-v2/CommandPolicy.tsapps/server/src/orchestration-v2/DelegatedCompletionDelivery.test.tsapps/server/src/orchestration-v2/EffectWorker.tsapps/server/src/orchestration-v2/IdleCompaction.test.tsapps/server/src/orchestration-v2/Orchestrator.tsapps/server/src/orchestration-v2/ProjectionStore.test.tsapps/server/src/orchestration-v2/ProjectionStore.tsapps/server/src/orchestration-v2/ProviderContinuationService.tsapps/server/src/orchestration-v2/ProviderSessionManager.test.tsapps/server/src/orchestration-v2/ProviderTurnStartService.test.tsapps/server/src/orchestration-v2/ProviderTurnStartService.tsapps/server/src/orchestration-v2/SteeringCompletion.integration.test.tsapps/server/src/orchestration-v2/ThreadLaunchService.tsapps/server/src/orchestration-v2/ThreadManagementService.tsapps/server/src/orchestration-v2/ThreadMessageIntake.tsapps/server/src/orchestration-v2/ThreadSettlementService.test.tsapps/server/src/orchestration-v2/ThreadSettlementService.tsapps/server/src/orchestration-v2/ThreadStop.test.tsapps/server/src/orchestration-v2/TurnItemPositionStore.test.tsapps/server/src/orchestration-v2/TurnItemPositionStore.tsapps/server/src/orchestration-v2/http.test.tsapps/server/src/orchestration-v2/runtimeLayer.test.tsapps/server/src/orchestration-v2/runtimeLayer.tsapps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.tsapps/server/src/orchestration-v2/testkit/ReplayRuntimeInstructions.tsapps/server/src/persistence/Sqlite.tsapps/server/src/provider/CodexProvider.test.tsapps/server/src/provider/CodexProvider.tsapps/server/src/provider/ProviderOrchestrationAdapterInfrastructure.tsapps/server/src/provider/agentContextBudget.jsonapps/server/src/provider/codexUsageLimits.test.tsapps/server/src/server.tsapps/server/src/storageCleanup.tsapps/server/src/usage/UsageLimitSources.tsapps/server/src/vcs/GitVcsDriverCore.tsapps/server/src/ws.tsapps/web/src/components/ChatView.logic.tsapps/web/src/components/ChatView.tsxapps/web/src/components/chat/ChatComposer.tsxapps/web/src/components/chat/ComposerControlsBar.tsxapps/web/src/components/chat/ComposerFooterControls.tsxapps/web/src/components/chat/ComposerPrimaryActions.tsxapps/web/src/components/chat/MessagesTimeline.browser.test.tsxapps/web/src/components/chat/MessagesTimeline.logic.test.tsapps/web/src/components/chat/MessagesTimeline.logic.tsapps/web/src/components/chat/TimelineAssistantRows.tsxapps/web/src/components/chat/composerCommandKey.tsapps/web/src/components/chat/composerProviderState.test.tsxapps/web/src/components/chat/useComposerSend.tsapps/web/src/components/settings/ScheduledTasksSettings.tsxapps/web/src/components/settings/scheduledTasksSettings.logic.tsapps/web/src/composerDraftStore.test.tsapps/web/src/hooks/useThreadActionMenu.tsapps/web/src/versionSkew.test.tsapps/web/src/versionSkew.tsdocs/user/background-service.mdevals/mcp-tool-use/run.test.tsevals/mcp-tool-use/run.tspackages/client-runtime/package.jsonpackages/client-runtime/src/state/commandPermissions.test.tspackages/client-runtime/src/state/orchestrationV2Projection.test.tspackages/client-runtime/src/state/threadExecution.tspackages/contracts/src/orchestrationV2.tspackages/contracts/src/orchestratorMcp.test.tspackages/contracts/src/orchestratorMcp.tspackages/provider-acp/src/server/adapter.tspackages/provider-core/src/server/usageLimits.tspackages/provider-opencode/src/server/OpenCodeServerLedger.test.tspackages/provider-opencode/src/server/OpenCodeServerLedger.tspackages/shared/src/orchestrationV2Timeline.test.tsscripts/dev-runner.tsImplemented by GPT-6.1 Sol, high effort, through the Codex harness (primary and three delegated implementation workers).
Alb11747 commentedon Nov 14, 2026
Fork sync resolution — request 43580534179
Candidate:
c1eda99e9fa9dcc5b6c03aed7f451172b9a3ca62. Fork base:51d334f8a2107185383cf45e6250e905e44762e9. Requested upstream:db2cc2026427c5eb647a08538c12402341db0fab.The initial merge stopped in 91 files. Resolved the merge while preserving upstream structure and retained fork policies. No webhook request data was used. Only the copied trusted publisher will publish this candidate; main was not pushed.
Superseded patches
Aborted the first merge and reverted these already-equivalent patches before merging again:
4414c245461db55d9700eca3af1656f142f909b2, inherited dev-server bind settings, replaced by upstreamf9cadccadf4f6d6e37b005b7fec143c70cadbd6c(fix(scripts): dev runner ignores the parent T3 server's port and host pingdotgg/t3code#36019).c3f2943bafcdf5ad683dde3870ae4619889ba654, invalid gitfile cleanup, replaced by upstream301999c325cebee9f4330ee14ef8c3b083e67a64(fix(server): git 2.56 broken-gitfile worktrees count as non-repositories again pingdotgg/t3code#33423).Removed their entries from
fork/CHANGES.md. Reverts are ordinary commits; existing fork history was retained. The remaining conflicts were resolved in merge4051939c72; follow-up adaptations are in49114600a1e31c683c32e7c73d99678c57ea58b5.Resolutions
Test merges distorted by upstream removing redundant Effect.scoped wrappers were aligned through temporary AST/scope normalization, retaining the actual test behaviors. Temporary merge/debug aids are outside the checkout; no debug code is committed.
Validation
corepack pnpm exec vp fmt --check: passed on all7380 matched files. Final diff/state/ancestry checks: clean checkout, no unresolved entries or merge/revert state, both requested parents are ancestors; diff whitespace findings verified upstream-only as detailed below.Earlier parallel checks coincided with the container's12GiB memory limit and host load, producing OOM exits and worker/time-out failures. Superseded owned runs were stopped; later tests used maxWorkers1. No timeouts, assertions or type checks were weakened. Actual semantic failures were reproduced and fixed; environment deaths are identified separately above.
git diff --checkreports whitespace only inside upstream-owned patch filespatches/@cursor__sdk@1.2.0.patchandpatches/@pierre__diffs@1.7.0-beta.3.patch. Their candidate blobs match the requested upstream blobs exactly (cf657be869d5c688ad99443f02097f1ac3d8559c,5704f9c799cf0dbd8ceb2486da1e8e49fe60f49c); left unchanged per procedure, not counted as failed validation.Review focus
Inspect the boundary between fork native steering and upstream deferred startup/Stop; every-turn versus settled child delivery; context-note versus newest-first timeline reads; streamed versus confirmed answer phases; and CLI ownership of fork schema additions. Upstream fixes native usage routing more broadly than the retained unsupported-read probe patch: please check whether that fork patch can now be fully retired. OpenCode's legacy/version2 record compatibility and wrapper fallback also deserve review.
The only resolver-authored change under
fork/isfork/CHANGES.md(two superseded entries removed and formatting). No resolver-authored workflow/action change; neither the procedure norfork/scripts/t3-resolver/was edited. All files in the conflict inventory follow.Hosted validation fix round 1
The first hosted run 43584558785 failed lint and focused web/provider/server cases. These are fixed in ordinary commit
c1eda99e9fa9dcc5b6c03aed7f451172b9a3ca62; the existing merge and initial adaptation commits are retained. Review PR: #92.Fix-round validation, all successful:
vp fmt --check(7380 files), and changed-content whitespace checks: passed. Only the previously documented exact-upstream patch whitespace remains in the full base-to-candidate comparison.All tests used
--maxWorkers=1. Owned test/typecheck processes finished before publication; no temporary tool/error logging remains. The earlier source/fixture report and initial conflict inventory below remain applicable.Initial conflict inventory
apps/desktop/src/app/DesktopApp.tsapps/desktop/src/settings/DesktopAppSettings.tsapps/desktop/src/updates/DesktopUpdates.tsapps/desktop/src/window/DesktopWindow.test.tsapps/desktop/src/window/DesktopWindow.tsapps/mobile/src/features/settings/SettingsScheduledTasksRouteScreen.tsxapps/mobile/src/features/threads/ThreadSettingsSheet.tsxapps/mobile/src/lib/threadActivity.tsapps/mobile/src/state/thread-outbox-model.tsapps/mobile/src/state/use-thread-composer-state.tsapps/server/src/mcp/OrchestratorMcpService.activity.test.tsapps/server/src/mcp/OrchestratorMcpService.test.tsapps/server/src/mcp/OrchestratorMcpService.tsapps/server/src/mcp/toolkits/orchestrator/handlers.tsapps/server/src/mcp/toolkits/orchestrator/tools.tsapps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.tsapps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.tsapps/server/src/orchestration-v2/Adapters/CodexAdapterV2.test.tsapps/server/src/orchestration-v2/Adapters/CodexAdapterV2.tsapps/server/src/orchestration-v2/ClaudeAutomaticDelivery.integration.test.tsapps/server/src/orchestration-v2/CommandPolicy.tsapps/server/src/orchestration-v2/DelegatedCompletionDelivery.test.tsapps/server/src/orchestration-v2/EffectWorker.tsapps/server/src/orchestration-v2/IdleCompaction.test.tsapps/server/src/orchestration-v2/Orchestrator.tsapps/server/src/orchestration-v2/ProjectionStore.test.tsapps/server/src/orchestration-v2/ProjectionStore.tsapps/server/src/orchestration-v2/ProviderContinuationService.tsapps/server/src/orchestration-v2/ProviderSessionManager.test.tsapps/server/src/orchestration-v2/ProviderTurnStartService.test.tsapps/server/src/orchestration-v2/ProviderTurnStartService.tsapps/server/src/orchestration-v2/SteeringCompletion.integration.test.tsapps/server/src/orchestration-v2/ThreadLaunchService.tsapps/server/src/orchestration-v2/ThreadManagementService.tsapps/server/src/orchestration-v2/ThreadMessageIntake.tsapps/server/src/orchestration-v2/ThreadSettlementService.test.tsapps/server/src/orchestration-v2/ThreadSettlementService.tsapps/server/src/orchestration-v2/ThreadStop.test.tsapps/server/src/orchestration-v2/TurnItemPositionStore.test.tsapps/server/src/orchestration-v2/TurnItemPositionStore.tsapps/server/src/orchestration-v2/http.test.tsapps/server/src/orchestration-v2/runtimeLayer.test.tsapps/server/src/orchestration-v2/runtimeLayer.tsapps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.tsapps/server/src/orchestration-v2/testkit/ReplayRuntimeInstructions.tsapps/server/src/persistence/Sqlite.tsapps/server/src/provider/CodexProvider.test.tsapps/server/src/provider/CodexProvider.tsapps/server/src/provider/ProviderOrchestrationAdapterInfrastructure.tsapps/server/src/provider/agentContextBudget.jsonapps/server/src/provider/codexUsageLimits.test.tsapps/server/src/server.tsapps/server/src/storageCleanup.tsapps/server/src/usage/UsageLimitSources.tsapps/server/src/vcs/GitVcsDriverCore.tsapps/server/src/ws.tsapps/web/src/components/ChatView.logic.tsapps/web/src/components/ChatView.tsxapps/web/src/components/chat/ChatComposer.tsxapps/web/src/components/chat/ComposerControlsBar.tsxapps/web/src/components/chat/ComposerFooterControls.tsxapps/web/src/components/chat/ComposerPrimaryActions.tsxapps/web/src/components/chat/MessagesTimeline.browser.test.tsxapps/web/src/components/chat/MessagesTimeline.logic.test.tsapps/web/src/components/chat/MessagesTimeline.logic.tsapps/web/src/components/chat/TimelineAssistantRows.tsxapps/web/src/components/chat/composerCommandKey.tsapps/web/src/components/chat/composerProviderState.test.tsxapps/web/src/components/chat/useComposerSend.tsapps/web/src/components/settings/ScheduledTasksSettings.tsxapps/web/src/components/settings/scheduledTasksSettings.logic.tsapps/web/src/composerDraftStore.test.tsapps/web/src/hooks/useThreadActionMenu.tsapps/web/src/versionSkew.test.tsapps/web/src/versionSkew.tsdocs/user/background-service.mdevals/mcp-tool-use/run.test.tsevals/mcp-tool-use/run.tspackages/client-runtime/package.jsonpackages/client-runtime/src/state/commandPermissions.test.tspackages/client-runtime/src/state/orchestrationV2Projection.test.tspackages/client-runtime/src/state/threadExecution.tspackages/contracts/src/orchestrationV2.tspackages/contracts/src/orchestratorMcp.test.tspackages/contracts/src/orchestratorMcp.tspackages/provider-acp/src/server/adapter.tspackages/provider-core/src/server/usageLimits.tspackages/provider-opencode/src/server/OpenCodeServerLedger.test.tspackages/provider-opencode/src/server/OpenCodeServerLedger.tspackages/shared/src/orchestrationV2Timeline.test.tsscripts/dev-runner.tsImplemented by GPT-6.1 Sol, high effort, through the Codex harness (primary and three delegated implementation workers).
Alb11747 commentedon Nov 14, 2026
Fork sync resolution — request 43580534179
Candidate:
3178120fcefbfb9c9436323a0da2124104b37da0. Fork base:51d334f8a2107185383cf45e6250e905e44762e9. Requested upstream:db2cc2026427c5eb647a08538c12402341db0fab.The initial merge stopped in 91 files. Resolved the merge while preserving upstream structure and retained fork policies. No webhook request data was used. Only the copied trusted publisher will publish this candidate; main was not pushed.
Superseded patches
Aborted the first merge and reverted these already-equivalent patches before merging again:
4414c245461db55d9700eca3af1656f142f909b2, inherited dev-server bind settings, replaced by upstreamf9cadccadf4f6d6e37b005b7fec143c70cadbd6c(fix(scripts): dev runner ignores the parent T3 server's port and host pingdotgg/t3code#36019).c3f2943bafcdf5ad683dde3870ae4619889ba654, invalid gitfile cleanup, replaced by upstream301999c325cebee9f4330ee14ef8c3b083e67a64(fix(server): git 2.56 broken-gitfile worktrees count as non-repositories again pingdotgg/t3code#33423).Removed their entries from
fork/CHANGES.md. Reverts are ordinary commits; existing fork history was retained. The remaining conflicts were resolved in merge4051939c72; follow-up adaptations are in49114600a1e31c683c32e7c73d99678c57ea58b5.Resolutions
Test merges distorted by upstream removing redundant Effect.scoped wrappers were aligned through temporary AST/scope normalization, retaining the actual test behaviors. Temporary merge/debug aids are outside the checkout; no debug code is committed.
Validation
corepack pnpm exec vp fmt --check: passed on all7380 matched files. Final diff/state/ancestry checks: clean checkout, no unresolved entries or merge/revert state, both requested parents are ancestors; diff whitespace findings verified upstream-only as detailed below.Earlier parallel checks coincided with the container's12GiB memory limit and host load, producing OOM exits and worker/time-out failures. Superseded owned runs were stopped; later tests used maxWorkers1. No timeouts, assertions or type checks were weakened. Actual semantic failures were reproduced and fixed; environment deaths are identified separately above.
git diff --checkreports whitespace only inside upstream-owned patch filespatches/@cursor__sdk@1.2.0.patchandpatches/@pierre__diffs@1.7.0-beta.3.patch. Their candidate blobs match the requested upstream blobs exactly (cf657be869d5c688ad99443f02097f1ac3d8559c,5704f9c799cf0dbd8ceb2486da1e8e49fe60f49c); left unchanged per procedure, not counted as failed validation.Review focus
Inspect the boundary between fork native steering and upstream deferred startup/Stop; every-turn versus settled child delivery; context-note versus newest-first timeline reads; streamed versus confirmed answer phases; and CLI ownership of fork schema additions. Upstream fixes native usage routing more broadly than the retained unsupported-read probe patch: please check whether that fork patch can now be fully retired. OpenCode's legacy/version2 record compatibility and wrapper fallback also deserve review.
The only resolver-authored change under
fork/isfork/CHANGES.md(two superseded entries removed and formatting). No resolver-authored workflow/action change; neither the procedure norfork/scripts/t3-resolver/was edited. All files in the conflict inventory follow.Hosted validation fix round 1
The first hosted run 43584558785 failed lint and focused web/provider/server cases. These are fixed in ordinary commit
c1eda99e9fa9dcc5b6c03aed7f451172b9a3ca62; the existing merge and initial adaptation commits are retained. Review PR: #92.Fix-round validation, all successful:
vp fmt --check(7380 files), and changed-content whitespace checks: passed. Only the previously documented exact-upstream patch whitespace remains in the full base-to-candidate comparison.All tests used
--maxWorkers=1. Owned test/typecheck processes finished before publication; no temporary tool/error logging remains. The earlier source/fixture report and initial conflict inventory below remain applicable.Hosted validation fix round 2 — final candidate
The second hosted run 43589689173 reported two remaining failing tests. Both were reproduced locally before changes, and corrected in ordinary commit
3178120fcefbfb9c9436323a0da2124104b37da0without changing the upstream regressions.85b160658f(fix(server): keep thread tool responses small by default pingdotgg/t3code#29268) removed it for500-character default/paginated reads;6237824a2c(fix(server): t3_thread_read results fit MCP clients' output limit pingdotgg/t3code#33455) retains empty/past-end safety. Both commits are ancestors of the requested upstream SHA. I incorrectly restored the obsolete condition in round1 based on a fixture that had actually read the entire result as a prefix plus remainder. The final source restores upstream's nonempty completed-page rule. The fork integration now verifies the complete reconstructed result is acknowledged and adds a strict check that an empty past-end read does not acknowledge it; truncated-prefix delivery stays pending. No read-coverage state, API, or upstream-test weakening was introduced.fb2e5a0128(fix(client-runtime): imported threads that never ran no longer sit in Working as Waiting pingdotgg/t3code#33042), also present in the requested merge, intentionally removed an active-provider guard from runtime derivation for imported/runless threads. The merge had accidentally retained that obsolete guard while preserving fork issue-watch input. The final source adopts the upstream condition and retains issue-watch derivation. A new focused control proves watched threads with no app runs remain idle/observable and become null when the watch is removed; the unchanged provider-native subagent regression now passes.Final-round validation:
All owned test/typecheck processes have finished. This is the second and final permitted fix round; promotion still requires the trusted helper's hosted checks and exact-commit Codex review.
Initial conflict inventory
apps/desktop/src/app/DesktopApp.tsapps/desktop/src/settings/DesktopAppSettings.tsapps/desktop/src/updates/DesktopUpdates.tsapps/desktop/src/window/DesktopWindow.test.tsapps/desktop/src/window/DesktopWindow.tsapps/mobile/src/features/settings/SettingsScheduledTasksRouteScreen.tsxapps/mobile/src/features/threads/ThreadSettingsSheet.tsxapps/mobile/src/lib/threadActivity.tsapps/mobile/src/state/thread-outbox-model.tsapps/mobile/src/state/use-thread-composer-state.tsapps/server/src/mcp/OrchestratorMcpService.activity.test.tsapps/server/src/mcp/OrchestratorMcpService.test.tsapps/server/src/mcp/OrchestratorMcpService.tsapps/server/src/mcp/toolkits/orchestrator/handlers.tsapps/server/src/mcp/toolkits/orchestrator/tools.tsapps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.tsapps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.tsapps/server/src/orchestration-v2/Adapters/CodexAdapterV2.test.tsapps/server/src/orchestration-v2/Adapters/CodexAdapterV2.tsapps/server/src/orchestration-v2/ClaudeAutomaticDelivery.integration.test.tsapps/server/src/orchestration-v2/CommandPolicy.tsapps/server/src/orchestration-v2/DelegatedCompletionDelivery.test.tsapps/server/src/orchestration-v2/EffectWorker.tsapps/server/src/orchestration-v2/IdleCompaction.test.tsapps/server/src/orchestration-v2/Orchestrator.tsapps/server/src/orchestration-v2/ProjectionStore.test.tsapps/server/src/orchestration-v2/ProjectionStore.tsapps/server/src/orchestration-v2/ProviderContinuationService.tsapps/server/src/orchestration-v2/ProviderSessionManager.test.tsapps/server/src/orchestration-v2/ProviderTurnStartService.test.tsapps/server/src/orchestration-v2/ProviderTurnStartService.tsapps/server/src/orchestration-v2/SteeringCompletion.integration.test.tsapps/server/src/orchestration-v2/ThreadLaunchService.tsapps/server/src/orchestration-v2/ThreadManagementService.tsapps/server/src/orchestration-v2/ThreadMessageIntake.tsapps/server/src/orchestration-v2/ThreadSettlementService.test.tsapps/server/src/orchestration-v2/ThreadSettlementService.tsapps/server/src/orchestration-v2/ThreadStop.test.tsapps/server/src/orchestration-v2/TurnItemPositionStore.test.tsapps/server/src/orchestration-v2/TurnItemPositionStore.tsapps/server/src/orchestration-v2/http.test.tsapps/server/src/orchestration-v2/runtimeLayer.test.tsapps/server/src/orchestration-v2/runtimeLayer.tsapps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.tsapps/server/src/orchestration-v2/testkit/ReplayRuntimeInstructions.tsapps/server/src/persistence/Sqlite.tsapps/server/src/provider/CodexProvider.test.tsapps/server/src/provider/CodexProvider.tsapps/server/src/provider/ProviderOrchestrationAdapterInfrastructure.tsapps/server/src/provider/agentContextBudget.jsonapps/server/src/provider/codexUsageLimits.test.tsapps/server/src/server.tsapps/server/src/storageCleanup.tsapps/server/src/usage/UsageLimitSources.tsapps/server/src/vcs/GitVcsDriverCore.tsapps/server/src/ws.tsapps/web/src/components/ChatView.logic.tsapps/web/src/components/ChatView.tsxapps/web/src/components/chat/ChatComposer.tsxapps/web/src/components/chat/ComposerControlsBar.tsxapps/web/src/components/chat/ComposerFooterControls.tsxapps/web/src/components/chat/ComposerPrimaryActions.tsxapps/web/src/components/chat/MessagesTimeline.browser.test.tsxapps/web/src/components/chat/MessagesTimeline.logic.test.tsapps/web/src/components/chat/MessagesTimeline.logic.tsapps/web/src/components/chat/TimelineAssistantRows.tsxapps/web/src/components/chat/composerCommandKey.tsapps/web/src/components/chat/composerProviderState.test.tsxapps/web/src/components/chat/useComposerSend.tsapps/web/src/components/settings/ScheduledTasksSettings.tsxapps/web/src/components/settings/scheduledTasksSettings.logic.tsapps/web/src/composerDraftStore.test.tsapps/web/src/hooks/useThreadActionMenu.tsapps/web/src/versionSkew.test.tsapps/web/src/versionSkew.tsdocs/user/background-service.mdevals/mcp-tool-use/run.test.tsevals/mcp-tool-use/run.tspackages/client-runtime/package.jsonpackages/client-runtime/src/state/commandPermissions.test.tspackages/client-runtime/src/state/orchestrationV2Projection.test.tspackages/client-runtime/src/state/threadExecution.tspackages/contracts/src/orchestrationV2.tspackages/contracts/src/orchestratorMcp.test.tspackages/contracts/src/orchestratorMcp.tspackages/provider-acp/src/server/adapter.tspackages/provider-core/src/server/usageLimits.tspackages/provider-opencode/src/server/OpenCodeServerLedger.test.tspackages/provider-opencode/src/server/OpenCodeServerLedger.tspackages/shared/src/orchestrationV2Timeline.test.tsscripts/dev-runner.tsImplemented by GPT-6.1 Sol, high effort, through the Codex harness (primary and three delegated implementation workers).
Alb11747 commentedon Nov 14, 2026
Candidate
3178120fcefbfb9c9436323a0da2124104b37da0was not promoted automatically: Checks finished with failure after 2 fix attempts. It needs human review.Alb11747 commentedon Nov 14, 2026
Maintainer follow-up for request 43580534179. Final candidate:
2476a907806269f7e97424613e4221bc98387116.The upstream merge, adaptations, and two resolver CI fix rounds remain intact; their detailed summary is #91 (comment) and the preceding resolution comments. The original candidate
3178120fcefbfb9c9436323a0da2124104b37da0passed all hosted jobs in run 43591599431 attempt 2 after retrying only the failed Web job. The initial browser connection loss remains unexplained; its named upstream-identical test also passed locally.One independent Opus review covered the high-risk resolutions and follow-up commits. The confirmed test gap is corrected in the final commit: the integration now checks that the persisted reservation matches the offered generation and message ID and contains both completed and interrupted follow-up task IDs. This adds coverage without changing production behavior or removing assertions.
The proposed startup waiting-run regression was dismissed after direct reproduction. Widening the read still reaches an explicit rejection in dispatchSteerIntoRun for any target status other than running. Both fork base 51d334f and pinned upstream db2cc20 have that guard, so the claimed formerly successful waiting-run steering did not occur. Experimental source/test diagnostics were fully removed; the only maintainer source diff is the nine-line coverage assertion.
Final local validation: check:changed against 3178120 passed format, lint, both dependency checks, GraphQL generation check, three package typechecks, and three tests in two files, under Node 26.12.1. git diff --check and LF checks passed. The complete MCP integration passed with the new task coverage. No live provider/client/remote session was exercised.
Hosted Codex review remains quota-blocked; no credits were added or automated review metadata changed. The independent review and maintainer findings disposition are recorded at #92 (comment). Fresh hosted validation is required on this final candidate before manual promotion. Main has not been pushed by the maintainer.
Posted by GPT-6.1 Sol on behalf of Alb11747.
github-actions commentedon Nov 14, 2026
Promoted to main by https://github.com/Alb11747/t3code/actions/runs/43596293290.