Repository navigation
Sync upstream main through ffd641575 - #186
Merged
Merged
Conversation
…rompt#1425) repoprompt#1346 removed client-name auto-allow; the bundled helper is now auto-approved only when clientInfo.name starts with "RepoPrompt CLI". The external initialize probe sent codex-mcp-client, waiting 300s for manual approval past the 60s watchdog. Rename the probe client, keep object-valued capabilities, and preserve the probe's stdout/stderr, exit status, and helper debug tail on failure. Add source-text regression tests.
…ff previews (repoprompt#1430) * test(grok): red replay of recorded 1.0.62 native edits (repoprompt#1327) * test(grok): embed recorded 1.0.62 fixtures as Swift raw strings (repoprompt#1327) * Grok Build: show native search_replace/write results as edit diffs (repoprompt#1327) * test(grok): self-contained fixture provenance; normalizer doc note (repoprompt#1327) * test(grok): reproduce native edit gutter offsets live and restored (repoprompt#1327) * Grok Build: preserve terminal edit line numbers in diff previews (repoprompt#1327) * Grok Build: bound split edit lines and cover canonical failure gate (repoprompt#1327) * Grok Build: share diff input limits from the ACP adapter (repoprompt#1327)
…mpt#1432) * feat(agent): persist scoped Re-prompt settings with schema 13 * feat(agent): add isolated Re-prompt rewriting and context consults * feat(agent): gate local composer turns on reviewed briefs and reuse automation judgments * feat(agent): add Re-prompt composer review and scoped guidance editor * docs: record Re-prompt validation results and remaining acceptance gaps * fix(agent-mode): address re-prompt review findings * feat(agent): stream Re-prompt preparation in composer Replace modal consult/review with claim-free Markdown streaming, inline clarification, explicit second-Send dispatch, independent subscription-first model selection, bounded recent-chat context, recovery and fenced speculative routing. Remove Re-prompt-only context consultation plumbing and extend schema 13 model settings. Record v2 design, deviations, privacy disclosure, file inventory and coordinated verification. * feat(agent): refine Re-prompt composer controls and modes Refresh cached pill activity at every phase transition. Add inline original/revert, scoped Cmd-Return repeat and one-shot Review, Escape/native-undo recovery, and opt-in profile-scoped auto-send with clarification fallback and fresh-claim fencing. Present concrete model choices, migrate legacy Oracle linkage, persist history preferences, collapse advanced guidance, and document v2.1 verification. * fix(agent): preserve Re-prompt intent and satisfy review gates * test(settings): pin global settings document at schema 13 for Re-prompt fields
…ge override (repoprompt#1433) Even pace scored plans by used/elapsed, which looks backward: Claude at 32% used with 37h to reset and Codex at 6% with 6 days both scored ~0.4, so balancing never moved work off the router's GPT pick while Claude allowance expired unused. It also treated a freshly reset plan as "behind". - Even pace now uses forward spare pace: unused share divided by the share of the weekly window left (tightest applicable window). Weekly readings stay usable for six hours instead of one. - New opt-in "Override Model Router when usage is critical". A plan is critical when it has >=25% weekly allowance resetting within 48h (even if the other plan's usage is unknown), 1.75x the other plan's spare pace, or the other plan is >=90% used. Only the dominant provider's models reach the decision service; a single remaining model skips it entirely. - Primary sessions under the override are floored at strong/frontier tiers (Opus/Fable or Sol/Astra) so trivial prompts never land on Haiku, Sonnet, Luna, or Terra; subagents and lanes keep every tier. - The no-router fallback applies the same override across tiers and efforts, keeping the existing paid Fast / parameterized / non-curated starting-target exclusions.
Reclaim scope-cache leases on retired shard objects before enforcing the live-shard cap. Preserve external snapshot readers and current-publication cache entries. Refs repoprompt#1416
…#1435) * fix(mcp): retain process-scoped approval across reconnects * fix(mcp): preserve FIFO approval order across actor suspension
* fix: bound app termination and sweep owned agent processes * Reduce quit-hang fix and run SIGTERM terminate from the run loop SIGTERM called NSApp.terminate(nil) from a main-queue dispatch-source callout. With .terminateLater, AppKit spins a nested run loop inside that callout, which holds the main queue, so the @mainactor work that sends the reply never runs. An idle app ignored SIGTERM for more than 15 seconds. Cmd+Q was unaffected. Perform terminate from the run loop instead. Shrink the termination bound to what the issue needs. The coordinator races shutdown against a 5 second deadline, runs a synchronous child sweep, and replies once. The registry is a pid set that SIGKILLs still-unreaped child process groups on app exit. Drop waitpid routing, launch admission, the poll loop, and the graceful-path sweep. * Trim quit fix to the SIGTERM run-loop change Fable deslop review: the reproduced bug is SIGTERM calling NSApp.terminate from a main-queue callout, which deadlocks the .terminateLater reply. The deadline coordinator and child registry addressed a hung await that was never reproduced, and graceful shutdown already bounds child teardown via beginAppTerminationFastPath. Remove them with their tests and the launcher hook.
…mits (repoprompt#1450) * fix(agent-mode): retain tool payloads across session saves * fix(agent-mode): preserve payloads during in-place transcript commits
…ompt#1439) * Fix saved oversight restoration after duplicate windows close * Address review feedback on duplicate-window oversight restoration * Keep duplicate-restoration notices synchronized with current presentation * Make duplicate-restoration notices edge-triggered and dismissible
* Fix snapshot publication test activation race * Pin snapshot test unpublished-selection precondition
…ledger row 129)
…witch send path
crstdr
marked this pull request as ready for review
October 23, 2026 10:41
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Sync fork main
65cb7a60awith upstream main pinned atffd64157596ff7ac4fccc3b3c8f9666a6c13bece, retaining fork-specific policies. The branch contains two merge commits, a separate dependency revert, and a bounded re-prompt/provider-switch integration repair; no rebase or cherry-picks.Upstream merges included
Local exclusions and integration
Behavior heads-up: inline re-prompt (repoprompt#1432), Grok native-edit previews (repoprompt#1430), quota balancing (repoprompt#1433), catalog shard leases (repoprompt#1436), SIGTERM exit (repoprompt#1169), retained tool-result preservation (repoprompt#1450), surfaced atomic-sync failures (repoprompt#1442), and saved-oversight restoration (repoprompt#1439) are included.
Validation
On
34d21dd816176b231e0c53e34ba855d355446a90:SessionAdminInventoryReadTests.testExactTargetApprovalAfterRequesterReconnectUsesProductionCatalogSessionAdminInventoryReadTests.testExactTargetApprovalContinuesColdSavedLinkAndManagedTaskWithoutUserTurnBoth match the existing Move sidebar workspace preparation off the main actor repoprompt/repoprompt-ce#1384-revert baseline: connection closed, missing
become_overseer, and an unwaited tools-list-change expectation. Tests were not weakened or newly skipped.RepoPromptproduct build: passed.windows(exit 73; no CE debug MCP connection). No visible app launch, stop or relaunch was performed.