RepoTunnel v0.4.1

RepoTunnel v0.4.1 is the first public 0.4.x release. It combines the capability work completed during the unpublished v0.4.0 validation build with the final Phone reliability fixes verified before release.

Highlights

• Added full Phone Access workflows with persistent scrcpy 4.1 live video, human control, semantic accessibility actions, bounded transactions, app/files/settings/network/shell tooling, and payment-sensitive privacy blocking for AI/MCP access.
• Fixed intermittent Phone Helper Resource temporarily unavailable (os error 11) failures by treating timed WouldBlock reads as transient on the same request/socket, with a bounded response deadline and no semantic-action replay.
• Fixed local Phone-panel lag by separating human tap/swipe/key/text control from the AI/MCP payment-policy helper path while keeping AI/MCP control on the strict payment-sensitive guard.
• Fixed stale/lagging Phone-panel video by giving the local human view a dedicated cached scrcpy frame path instead of running semantic-helper policy checks on every refresh; healthy rendering now waits briefly for the next frame instead of busy-polling duplicates.
• Treats empty semantic-helper socket closures as transient helper unavailability rather than protocol corruption, allowing safe read-only policy checks to recover cleanly.
• Added bounded large-project inspection/search/range-read tools with resumable cursors and concurrency protection for very large repositories.
• Strengthened long-running work with durable managed-process supervision and Linux restart reattachment backed by persisted supervisor state and process-identity checks.
• Added browser mutation receipts for selector/semantic click and type operations so post-dispatch transport loss is surfaced as ambiguous and never blindly replayed.
• Expanded Continuity/Resume, ChatGPT continuation recovery, Team Mode coordination, AI Workspace recovery, and current project documentation for long multi-session AI work.
• Expanded Video Production with persistent Video Projects, narration/subtitles, generated scenes, rendering/preview/QA, story adapters, and project-owned asset/license tracking.
• Final v0.4.1 release validation passed 43/43 frontend tests and 448 Rust tests with 0 failures and 8 intentional ignores.

RepoTunnel v0.3.1

RepoTunnel v0.3.1 is a focused reliability and editor-quality update for v0.3.0.

Highlights

• Fixed Projects / Continuity layouts so long live context cannot stretch the application horizontally or distort the page.
• Restored normal editor long-line behavior with visual line wrapping, while preserving the file contents unless the user explicitly inserts a newline.
• Added editor regression coverage for Backspace, undo/redo, caret stability, and long-line wrapping.
• Reduced UI stalls by moving heavier status/history/workspace operations off the frontend command path and coalescing repeated activity refreshes.
• Reduced AI Workspace preview polling pressure and cleaned up stale isolated-session processes more reliably.
• Cached external connection-tool probes so routine connection-status refreshes do not repeatedly spawn expensive checks.

RepoTunnel v0.3.0

RepoTunnel v0.3.0 focuses on reliable continuation, safe updates, and a hardened Direct HTTPS connection.

Highlights

• Added Continuity, which resumes from live Git, activity, process, and bounded project context instead of trusting stale saved next steps.
• Added signed in-app Auto Update infrastructure and release artifacts, with install-safety checks and persisted-state health verification.
• Fixed Direct HTTPS startup under Rustls 0.23 by selecting a deterministic crypto provider before TLS initialization.
• Made Direct HTTPS listener status truthful: :43183 is reported online only after TLS and ACME listeners initialize successfully.
• Kept the MCP/Direct HTTPS recovery channel independent from optional product features so normal connection startup cannot be stranded by unrelated UI features.
• Prevented Ollama Model Hub recovery from triggering privileged system-service password prompts; only the per-user service is attempted.
• Strengthened Project Memory/continuity persistence and connector safety metadata.

RepoTunnel v0.2.0

RepoTunnel v0.2.0 focuses on reliability, AI Workspace, editor quality, safety, and release hardening.

Highlights

• Added isolated AI Workspace automation with bounded multi-action sequences while preserving workspace, credential, self-control, browser, Docker, and teardown protections.
• Added platform-aware productivity integrations and verified VS Code with its integrated terminal as the practical Linux terminal workflow.
• Replaced the custom editor input layer with CodeMirror 6, including native undo/redo, selection, Backspace behavior, search, indentation, line operations, diagnostics, and lazy language support.
• Reduced background monitoring cost with metadata-only scans and moved heavy inspection/check work away from the desktop UI path.
• Added persistent Pause/Resume AI access, expanded Safety Scan coverage, request-grouped reversible history, checkpoints, and restore protections.
• Embedded the managed ngrok connection path with stable endpoint reuse, health controls, reconnect handling, and local credential protection.
• Added an advanced Direct HTTPS path with OAuth/MCP route allowlisting, trusted TLS, a loopback-only raw MCP gateway, and a documented Route64/WireGuard setup for CGNAT environments.
• Preserved fail-closed command isolation: Bubblewrap on Linux, AppContainer + Job Object on Windows, and Seatbelt compatibility sandboxing on macOS.
• Cleaned stale development artifacts and strengthened release checks for formatting, Clippy warnings, tests, dependency audits, version consistency, and package hygiene.
• Added release packaging for Linux DEB/RPM/AppImage, Windows NSIS/MSI, and macOS Apple Silicon/Intel DMG builds with SHA-256 checksums.

RepoTunnel v0.1.0

First public RepoTunnel release.

Linux packages:

• Debian/Ubuntu/Mint (.deb)
• Fedora/RHEL-compatible (.rpm)
• Portable Linux AppImage

Security highlights:

• Approved-workspace filesystem boundary
• Bubblewrap-isolated AI terminal execution
• Sensitive-file and secret protection
• OAuth-protected public MCP access
• Revocable MCP authorization
• Path traversal, symlink escape, and absolute-path protections
• Strict AI-directed temporary/runtime write restrictions
• Protected Git metadata and explicit push intent

Release validation:

• 58/58 Rust tests passed
• Frontend production build passed
• npm production audit: 0 vulnerabilities
• DEB, RPM, and AppImage privacy scans passed
• Installed DEB verified end-to-end with ChatGPT

See SHA256SUMS.txt for artifact checksums.