RalphX.app v0.104.0

RalphX.app 0.104.0 improves automated repair recovery, GitHub review resilience, and plan-approved UI clarity.

User-Facing Changes

• Repair-blocked workspaces now automatically resume when GitHub reports the current PR head fully green—no failing or in-flight checks—even when the head commit has not changed. d36f703a
• Stuck repair pauses now recover from provably unstarted push effects, and an age-bounded watchdog settles already-escalated attempts after the configured limit (default: six hours), allowing Retry publication to resolve previously wedged pauses. 0ec99cc9
• Workspace review actions now report GitHub rate-limit exhaustion with actionable retry-after-reset guidance rather than an opaque failure; archived workspaces with verified merged or closed PRs also stop spending repeated GitHub API calls. 2c8f951d
• Approved-plan banners with Tasks disabled now show only the “Plan approved” title and available actions, without the removed “Tasks is off” instruction or fallback description. c38e5140

Fixes And Polish

• Publication hold cards now prioritize curated, human-readable hold reasons and place raw diagnostics in an expandable Technical details section. 0ec99cc9

Developer And Maintainer Changes

• PR reconciliation now records verified publication associations, backfills terminal rows, and removes duplicate merged/closed publication events while sharing GitHub rate-limit state across reconciliation and polling paths. 2c8f951d

What's Changed

• Remove Tasks off message from plan approved widget by @lazabogdan in #1045
• feat: Supersede repair-blocked on PR mergeability by @lazabogdan in #1044
• Fix stuck conflict repair pause: inert effect classification and wedged-attempt watchdog by @lazabogdan in #1046
• Converge archived-workspace PR reconciliation and harden GitHub rate limiting by @lazaflorin in #1049

Full Changelog: v0.103.0...v0.104.0

RalphX.app v0.103.0

RalphX.app v0.103.0 reworks project setup and repository workflows, expands durable agent collaboration, and makes Workspace Review and publishing recovery more resilient.

User-Facing Changes

• Project setup and repository workflows now start with Clone Repository, Create New Repository, or Add Existing Repository; cloning shows live progress, supports cancellation, and preflights folders before registration. Local-only repositories are first-class and use Commit locally when PR publishing is unavailable. 281f721f 6010816e
• Workspace Review now opens confirmation sooner, supports per-conversation Auto Review & Fix overrides and capped fixer cycles, and uses prior settled reviews plus newly changed files for incremental re-review. A timed-out review with current final evidence can settle in a degraded state without arming auto-merge. aede223b a65794dd a3e65c70
• Publishing, base updates, and repair now show durable stage-by-stage progress, recover eligible repairs across restarts and base changes, and offer a confirmed Reopen PR flow that restores the local branch and worktree. 7c983479 4a656dcf c6cb37f5
• Delegation and Team Mode now track bound assignments in delegate-local ledgers, let coordinators wait or park until delegates settle, and provide feature-gated named teammates with durable context and coordinator wake-ups. 71bee4e5 6207aeec 72c7435a 001be358
• Planning workflows now use paired Overview and Blueprint artifacts, and Implement Directly performs a real Plan-to-Edit handoff on the first click. a3878d9a d7d31b3e 48a75fec
• Agents inbox and PR review navigation now organize conversations into stable Recent, Stale, and Done views, while PR Review conversations have their own lane with explicit states such as Approved. a6477799 57024970
• Jira and Confluence workflows now provide richer ticket context—including acceptance criteria and recent comments—and expand agent tools for boards, sprints, discovery, and markdown-based writes with role-scoped Atlassian permissions. 69839419 bc65d7df 10b1a1b2

Fixes And Polish

• Chat reliability and readability improve with chronological mid-run messages, durable queued-message recovery, grouped thinking blocks, live turn attribution, and stable bottom-follow scrolling with a pinned composer. 15adaec9 98d454e9 9768cdf0 5466f1a0
• Database and log maintenance now exposes configurable payload retention and opt-in size budgets with a confirmation preview; logs retain their newest output through capped rotation. e6d453e1 dad5f13d
• PR description publishing preserves reviewer-editable content on existing PRs, and description-generation failures no longer prevent eligible workspace publication. 3cb8a72a 54353733

Developer And Maintainer Changes

• Stable release promotion now resolves the prior Stable baseline, validates combined release notes before mutation, and supports a no-mutation notes dry run with artifact output. b4b0cf7b
• Backend composition now uses a dedicated shell layer with stronger import boundaries between Tauri startup wiring and application logic. 19a563c6

Known Issues

• Team members currently share a workspace, and automatic member-targeted wakes are not yet implemented; members learn about assignments through coordinator broadcasts or roster reads. 72c7435a 001be358
• Role-tiered Atlassian access applies to RalphX-spawned agents and RalphX-native delegates, not Claude native Task subagents; existing runs must be respawned to receive access. 10b1a1b2

What's Changed

• Implement Commit & Publish header alignment by @adriandemian in #874
• Safe Existing-PR Description Publishing Recovery by @lazabogdan in #890
• Native Delegate Assignments with Delegate-Local Task Ledgers by @lazabogdan in #888
• Reliable Agent-to-Plan Runtime Handoff by @lazabogdan in #886
• Add phase timing logs for review dialogs by @lazabogdan in #891
• Add delegation hints and repo rules by @lazabogdan in #893
• Add pinned Claude Opus 4.7, 4.8, and 5 controls by @lazabogdan in #892
• Support local-only Git projects with local commit and PR gating by @lazabogdan in #887
• Add paired Overview/Blueprint artifacts to planning and review flows by @lazabogdan in #889
• Fix RX-native Team mode prompt authority and empty-completion diagnosability by @lazabogdan in #896
• Test agent chat hydration consistency by @lazabogdan in #898
• Promote all Jira refs to durable context by @lazabogdan in #900
• PR Describer Publish False-Failure Recovery by @lazabogdan in #897
• Redesign diff hunks review filter as a walkthrough stepper by @lazabogdan in #902
• Redesign automation listing with priority filters by @adriandemian in #906
• Prevent cross-conversation plan artifact leakage by @adriandemian in #907
• Test and fix IPR message delivery by @lazabogdan in #903
• feat: add inbox-style agents sidebar with mute support by @lazabogdan in #909
• Redesign automations page with design tokens by @adriandemian in #911
• Fix missing artifact IDs after plan handoff by @lazabogdan in #910
• Surface automatic base-update repair progress with durable maintenance operations by @lazabogdan in #901
• Fix automation run cancellation bug by @adriandemian in #912
• Polish automation run tab UI to match design by @adriandemian in #913
• Fix live-stream chat rehydration ordering and segment concatenation bugs by @lazabogdan in #914
• Fix agents sidebar visibility and frontend error diagnostics by @lazabogdan in #915
• Refine inbox sidebar with dynamic labels by @lazabogdan in #920
• Automate stalled repair recovery in reconciliation loop by @lazabogdan in #916
• Refactor settings UI: consolidate nav to 7 sections, add integrations hub and search by @adriandemian in #918
• Fix agent task ledger isolation by @lazabogdan in #922
• Fix delegation context avail...

RalphX.app v0.102.0

RalphX.app 0.102.0 separates PR reviews in the inbox, fixes mid-run chat ordering, and makes Workspace Review more resilient and incremental.

User-Facing Changes

• PR Review inbox: Review PR conversations now use a dedicated PR Reviews lane with Needs you, Working, and Watching groups; rows show actual states such as Approved rather than a generic publish action. 57024970
• Workspace Review timeout recovery: When a reviewer writes a current final artifact but times out before completion, RalphX can settle the review gate from that typed evidence and explains the degraded settlement in the Review tab. Degraded outcomes do not arm auto-merge, so publishing remains manual. a3e65c70
• Incremental Workspace Review: Re-reviews now receive the prior settled review plus files changed since it, allowing them to re-check prior blocking or fold-in findings without restarting from the full delta when the history is available. a3e65c70
• Review context: Inline review excerpts now omit low-signal files such as lockfiles, generated output, snapshots, assets, and binaries while retaining on-demand diff pages; the excerpt budget increases from 42k to 60k. a3e65c70
• Commit & Publish annotations: Hunk annotations carry forward for files whose patch has not changed, leaving subsequent review cycles to annotate only newly changed hunks. a3e65c70

Fixes And Polish

• Live chat transcripts: User messages sent while an agent is running now appear at their chronological position between streamed content, and keep that ordering when the assistant message finalizes. 9768cdf0
• Sidebar navigation: The four inbox chips collapse to icon-and-count presentation at narrow sidebar widths to avoid label truncation. 57024970

Developer And Maintainer Changes

• Workspace Review operations: Reviewer deadlines now come from runtime configuration, while durable settlement evidence, annotator identity, and per-file annotation hashes are persisted through an additive migration. a3e65c70
• CI stability: Rust library test archive builds now cap Cargo concurrency at two jobs to avoid hosted-runner memory exhaustion. 57024970

What's Changed

• Optimize workspace review: separation of concerns, durable settlement, and incremental re-review by @lazabogdan in #1038
• fix: interleave mid-run user messages at their correct timeline position by @lazabogdan in #1030
• Separate PR review conversations into their own inbox lane with review-state by @lazaflorin in #1043

Full Changelog: v0.101.0...v0.102.0

RalphX.app v0.101.0

RalphX.app 0.101.0 expands Jira/Confluence workflows, improves unattended workspace-repair recovery, and corrects key publish and diff UI behavior.

User-Facing Changes

• Jira and Confluence workflows expand with board/sprint context, comment and user discovery, Confluence space discovery, markdown-based writes, and distinct composer references for Jira boards and Confluence links. bc65d7df
• Claude plan conversations now launch Implement Directly on the first click when a manual runtime override is selected, eliminating the previous double-click workaround. 48a75fec
• Workspace Review fixers can complete successfully without repair-authorization errors; reported blockers now appear on the review gate with their full detail while Fix Issues remains available for retry. 49f24524
• Published-workspace repair recovery now re-drives eligible blocked repairs after a base update, including CI-only needs_human holds that can be superseded by an automated base merge and CI restart; PR-autofix base updates also redrive unpublished heads without depending on agent completion classification. b1ddd939 48d65b4d 8d561be2
• Changes panel diffs no longer show base-branch progress as inverted workspace changes; workspaces with an invalid captured base fall back to the merge base at read time. 1cf2a3bc

Fixes And Polish

• Maintenance publish controls now use a shared availability gate: a zero-change repair can use Resume publish, Workspace Review blocks show their state, and disabled retry/resume actions explain why via tooltips. e23686ea
• Runtime stability on large or busy installations improves through debounced sidebar invalidation, lazy recovery-service construction, 250-row retention scans, and compaction that can proceed for oversized databases when disk-headroom safeguards allow it. Missing local worktrees now settle as recoverable Missing workspaces instead of repeatedly blocking recovery passes. 7f332e23

Developer And Maintainer Changes

• CI dependency setup now bounds apt index refreshes and Playwright system-dependency installation, allowing stalled mirrors to fall back to runner-image indexes rather than consume entire job timeouts. 49f24524

What's Changed

• fix: prod remediation — runtime flood, compaction gate, retention scan, orphaned worktrees by @lazabogdan in #1035
• Auto-redrive PR-autofix publishing after a base update, close redrive guard gap by @lazaflorin in #1037
• Fix silent no-op Resume publish button in maintenance banner by @lazabogdan in #1039
• Close Jira/Confluence integration gaps: agile, discovery, markdown writes, richer references by @lazaflorin in #1033
• Add unattended blocked-repair base-retry lane for published agent workspaces by @lazaflorin in #1036
• Fix Claude First-Click "Implement Directly" Failure by @lazaflorin in #1018
• Add completion channel for Workspace Review fixers by @lazabogdan in #1042
• fix: let base staleness supersede repair-blocked workspaces with CI-only holds by @lazabogdan in #1041
• Fix branch diff showing phantom changes from premature base retargeting by @lazabogdan in #1040

Full Changelog: v0.100.1...v0.101.0

RalphX.app v0.100.1

RalphX.app 0.100.1 makes Workspace Review deadlines liveness-aware, preserves completed reviews awaiting confirmation, and reports timeout failures more accurately.

User-Facing Changes

• Workspace Review now continues waiting while a reviewer is actively streaming timeline output, rather than ending a long review solely because a fixed 15-minute timeout elapsed; the default idle window is 10 minutes, with a one-hour wall-clock cap. 4b7ff100
• Review completion gets a 120-second grace period when a current Review artifact pair already exists, allowing complete_workspace_review_run to land instead of orphaning an otherwise finished review behind a failed gate. 4b7ff100

Fixes And Polish

• Workspace Review failures now distinguish between no Review being written, a Review awaiting completion confirmation, and a deadline expiring while the review monitor cannot be read. 4b7ff100
• Run-poll outages no longer leave the review waiter looping indefinitely: the wall-clock cap is evaluated even when the child-run record cannot be read. 4b7ff100

Developer And Maintainer Changes

• Workspace Review deadlines are now owned by the workspace_review runtime configuration, with matching RALPHX_* environment overrides for idle, wall-clock, and completion-grace limits. 4b7ff100
• Expanded focused regression coverage for monitor-read failures, run-poll failures, and CI rerun parsing edge cases such as empty OIDs and missing details URLs. 4b7ff100

What's Changed

• Fix workspace review timeout race condition with liveness-aware deadlines by @lazabogdan in #1020

Full Changelog: v0.100.0...v0.100.1

RalphX.app v0.100.0

RalphX.app 0.100.0 improves diagnostic log retention, GitHub-backed workspace recovery, and rate-limit resilience.

User-Facing Changes

• Diagnostic logs now retain the newest output when a per-launch size cap is reached: logs rotate into active and _rolled files instead of discarding later lines, while keeping total disk use within the configured cap. dad5f13d
• GitHub workspace automation now batches PR snapshots, adapts polling under rate-limit pressure, and defers automatic blocked-repair retries until the GitHub limit resets; rate-limit blocker copy now states that retry is automatic. e78ef6bd
• Orphaned PR-creation repairs now reconcile against GitHub: a confirmed absent PR clears the repair fence for retry, while a PR at the expected head is adopted and settled. af3ca5ff
• Interrupted repairs behind an advanced base now retarget to the current base and dispatch a successor instead of remaining blocked; repair banners also avoid exposing internal error variant names. e76bba5e

Fixes And Polish

• Data retention and compaction controls keep accessible names and busy states while operations run, including the cleanup dialog action; compaction swap failures now report the database’s actual surviving location. 5ed2efa6

Developer And Maintainer Changes

• Agent task-ledger context now renders an explicit empty marker and directs agents to trust the injected turn-start snapshot, avoiding redundant ledger reads unless the snapshot is missing, unavailable, or stale. 95891603
• Rust coverage CI restores four parallel library-coverage shards, removes unconsumed per-shard HTML reports, and runs coverage lanes when their workflow changes. 8ba5943a

What's Changed

• fix: preserve log tail by rotating on size cap instead of suppressing by @lazabogdan in #1022
• fix: data retention and compaction accessibility and error handling by @lazabogdan in #1023
• GitHub API Rate-Limit Hardening — PR Polling, Caching, and Transient Retry by @lazabogdan in #1027
• Optimize Rust lib coverage CI performance by @lazabogdan in #1031
• feat: Auto-reconcile orphaned in-flight create_pr repair effects by @lazabogdan in #1029
• Fix redundant agent task ledger tool calls by @lazabogdan in #1032
• Repair recovery loop fixes: dispatch-head backfill, base retarget, error detail isolation by @lazabogdan in #1034

Full Changelog: v0.99.0...v0.100.0

RalphX.app v0.99.0

RalphX.app 0.99.0 expands role-scoped Atlassian agent access, improves auto-repair recovery and hold visibility, and stabilizes desktop notifications.

User-Facing Changes

• Atlassian integration now gives RalphX-spawned agents direct Jira and Confluence MCP tools using the existing validated integration; per-role permissions in Settings → Agents → Roles offer Role default, None, Read, and Read + write access. 10b1a1b2
• Auto-repair holds now use a redesigned card with a status strip, clear headline, repair-agent narrative, primary action, and expandable Details timeline; transient base-parity CI holds offer Re-run failed checks. 4906c9ba
• Workspace publishing and repair no longer fail solely because PR-description generation fails: existing PR metadata is preserved, while new PRs use programmatic title and managed-body fallbacks. 54353733
• Desktop notifications now cap concurrent actionable click-waits and clear stale RalphX notifications after 15 minutes, preventing long unattended sessions from starving app responsiveness. 45a5b9b0

Fixes And Polish

• Blocked repair retries can now recover when an orphaned PR-update handoff has a durably observed branch push, clearing the effect fence so retry evaluation can continue. 2452359e
• Conflict-routed repairs no longer treat an unverified target base commit as integrated, preventing base-freshness workflows from being incorrectly parked. 54353733

Developer And Maintainer Changes

• Rust coverage CI timeouts were extended to accommodate normal shard runtime variation while retaining per-test hang detection. 4906c9ba 54353733

Known Issues

• Atlassian tools apply to RalphX-spawned agents and RalphX-native delegates; Claude native Task subagents do not receive role-tiered Atlassian access. Runs started before this update must be respawned for access. 10b1a1b2
• When more than three actionable notification click-waits are pending, additional notification clicks focus RalphX but do not deep-navigate until a slot is reclaimed. 45a5b9b0

What's Changed

• feat: Add built-in Atlassian MCP server with role-tiered access by @lazaflorin in #1019
• Fix blocked repair retries by terminating orphaned PR-update effects by @lazabogdan in #1021
• Auto-repair hold classification fixes and hold card UX redesign by @lazaflorin in #1016
• Fix main-thread starvation from unbounded desktop notification waits by @lazabogdan in #1026
• Decouple worktree repair from PR description failures by @lazaflorin in #1025

Full Changelog: v0.98.0...v0.99.0

RalphX.app v0.98.0

RalphX.app 0.98.0 makes chat-payload retention configurable, adds consent-backed size budgeting, and improves large-database compaction.

User-Facing Changes

• Data Retention in Settings → Database now exposes editable active and archived payload-retention windows (90 and 7 days by default) plus batch-size controls. e6d453e1
• Optional payload size budgets let you cap stored tool-call data; enabling a budget or reducing it shows a preview of affected rows, bytes, and cutoff time before confirmation. Message text, previews, timelines, and conversation metadata remain intact. e6d453e1
• Retention cleanup now runs continuously after startup and every six hours, so long-running installations no longer rely solely on restart-time pruning. Size-based cleanup remains off by default; a recommended 5 GB value is only a prefill for pristine installs. e6d453e1
• Database maintenance and startup now show a dedicated “Reclaiming disk space” stage, surface the latest compaction outcome and headroom warnings, and reduce compaction headroom from 3× the database size to roughly the compacted database size. e6d453e1

Fixes And Polish

• Database maintenance status no longer reports “0 B stored” when payload bytes were not measured; the size segment is omitted instead. e6d453e1

Developer And Maintainer Changes

• Backend composition now lives in a dedicated shell layer with enforced import boundaries, keeping Tauri-specific startup and wiring concerns separate from application logic. 19a563c6
• Layer ownership and guardrails were strengthened by relocating execution state, agent spawning, integration types, and ideation-apply logic to their appropriate layers, with updated CI paths and runtime-wiring coverage. 19a563c6

What's Changed

• Data Retention Overhaul — time-based + opt-in size budgeting by @lazabogdan in #1017
• refactor: separate shell layer and descend architectural types by @lazabogdan in #1015

Full Changelog: v0.97.0...v0.98.0

RalphX.app v0.97.0

RalphX.app 0.97.0 improves unattended workspace repair, closed-PR recovery, and publish workflow reliability.

User-Facing Changes

• Workspace drift and repair: Unpublished Edit workspaces now detect when their base branch moves ahead, show a warning indicator in the Agents sidebar, and can update from base when the agent is idle and existing auto-publish settings allow it. 4a656dcf
• Recovery across restarts: Unsettled workspace repairs receive startup recovery priority rather than being skipped by the capped PR-polling pool; durable repair reconciliation also covers non-GitHub projects. 4a656dcf
• Closed workspace PRs: The publish panel now offers a two-step Reopen PR flow that confirms live GitHub state, reopens closed PRs with explicit confirmation, restores the local branch and worktree from origin/<branch>, and restarts polling. c6cb37f5

Fixes And Polish

• PR descriptions: A successful PR-describer run with no submission now preserves an existing PR description instead of reporting an infrastructure failure; new PRs still require an explicit submission. d742d8a6
• Blocked repair controls: The publish panel now hides retry and publish actions when the backend determines blocked maintenance has no recoverable action. d742d8a6
• Operation feedback: Terminal workspace-operation toasts, including PR reopen results, now appear immediately instead of potentially waiting for the next polling interval. c6cb37f5

Developer And Maintainer Changes

• Reconciliation configuration: The unattended workspace repair scan defaults to a 60-second interval and can be configured through agent_workspace_repair_reconciliation_scan_interval_secs. 4a656dcf

What's Changed

• feat: Unattended Workspace Flow — base-ahead detection, repair reconciliation, startup prioritization by @lazaflorin in #1009
• feat: Reopen closed agent workspace PRs with local restore by @lazaflorin in #1012
• Fix PR describer error on existing PRs and gate repair retries by @lazabogdan in #1002

Full Changelog: v0.96.2...v0.97.0

RalphX.app v0.96.2

RalphX.app 0.96.2 fixes queued-message recovery and improves correlated completion delivery for workspace refresh and eligible automation.

User-Facing Changes

• Queued Messages no longer reappear after prompts were delivered and answered by an interactive agent; finalized assistant turns now settle the full delivered burst, while recovery suppresses turns with subsequent assistant activity. 083d45ac
• Completion-driven workspace refresh and eligible automation now use correlated, deduplicated run and turn completion events, preventing duplicate downstream completion handling during dual event delivery. c231ac3a

Developer And Maintainer Changes

• Chat startup, recovery, and workspace reconciliation now use explicit runtime dependencies and a shared event sink rather than direct managed-state or app-handle lookups. c231ac3a
• Added centralized completion dispatch and event catalog entries for agent:run_completed and agent:turn_completed, with bounded layering rules around completion producers. c231ac3a

What's Changed

• Fix queued messages re-appearing after send by @lazabogdan in #989
• Locator Removal and Correlated Event-Bus Cutover by @lazabogdan in #993

Full Changelog: v0.96.1...v0.96.2