Releases: roborev-dev/roborev
v0.54.1
v0.54.0
-
Add
roborev fix --batch-size Nfor processing findings in configurable batches. -
Add
roborev fix --resumefor continuing prior agent sessions. -
Keep large diff prompts external to handle bigger reviews more reliably.
-
Improve review prompt formatting so findings stay clearly separated.
-
Use a dark-background logo variant for GitHub dark mode.
-
Abort foreground fix runs cleanly when an agent hits quota or session limits.
-
Expose Codex diff snapshot directories for easier inspection.
v0.53.1
-
Improve ACP session validation before agent runs start.
-
Resolve ACP models more consistently across review, analyze, fix, refine, and daemon workflows.
-
Show clearer errors for invalid ACP configuration and model settings.
-
Improve ACP permission handling during session setup.
-
Fix branch base auto-detection when upstreams are configured as direct URLs.
v0.53.0
-
Adds an opt-in automatic design-review router for review workflows.
-
Adds smarter review type detection for routing code and design reviews.
-
Uses generated daemon API clients backed by the Huma/OpenAPI API surface.
-
Improves CI-triggered auto-design review handling.
-
Expands shell completion support for review type flags.
-
Improves repo and branch metadata handling across CLI, daemon, and TUI views.
-
Fixes
roborev fixreview discovery when running from a detached HEAD. -
Fixes TUI filtering after repositories are renamed.
-
Uses the branch upstream instead of the default branch for
--branchreviews. -
Completes review type flag handling across relevant commands.
-
Ignores Claude Code scheduled task files.
v0.52.0
-
Add Claude Code proxy support via
<model>@<base_url>configuration. -
Make review prompts more consistent and reduce low-value noise in review output.
-
Normalize streamed tool-call names and input fields for cleaner agent output.
-
Improve OpenCode output by showing tool calls and removing migration noise from stderr.
-
Fix TUI clipboard copy over SSH by using OSC52 escape sequences.
-
Let
j,k, andqbe typed normally while editing TUI filter searches. -
Fix Gemini severity threshold parsing when markers include internal whitespace.
v0.51.0
New Features
- Add OpenAPI support for the daemon API, including schema-driven endpoints for integrations and tooling.
- Add cascading
min_severityhandling and review-level filtering so you can focus on the issues that matter most.
Improvements
- Include per-commit review results in branch review prompts to give branch reviews better context.
- Include user comments and prior tool attempts in fix prompts so follow-up fixes are more informed.
- Honor global reasoning defaults more consistently across review, fix, refine, and enqueue flows.
- Allow viewing a job’s prompt even while the job is still queued.
- Improve Codex read-only sandbox support, including a unified diff file fallback for oversized diffs.
Bug Fixes
- Fix
roborev fixselecting jobs from unrelated branches. - Fix
roborev fixfinding open jobs when run from themainbranch. - Fix Codex sandbox invocation and command-line handling for better compatibility.
- Fix enqueue, rerun, config resolution, worktree, and CI edge cases that could cause incorrect job behavior.
- Fix GitHub Actions release checksum handling to use the correct
SHA256SUMSfilename. - Fix CI repo matching to resolve ambiguous matches instead of failing immediately.
v0.50.0
New Features
- Add
auto_close_passing_reviewsto automatically close passing reviews. - Add bundled
roborev-refineskills for Codex and Claude to support iterative review-fix-review loops. - Add systemd service/socket units and support systemd socket activation for packaged daemon deployments.
Improvements
- Update the TUI instantly by subscribing to the daemon event stream.
- Include human PR discussion in CI review prompts for better review context.
- Prefer
$XDG_RUNTIME_DIRfor the daemon socket path when it is available.
Bug Fixes
- Preserve requested model provenance when rerunning reviews.
- Prevent CI PR comment batches from hanging indefinitely by enforcing a timeout.
v0.49.0
-
Add daemon-backed
roborev insightsjobs so insights can be queued and tracked like reviews. -
Add Unix domain socket support for CLI-to-daemon communication on Unix systems.
-
Add
ROBOREV_COLOR_MODEto forceauto,dark,light, ornonecolor output. -
Improve skill installation and status reporting with a shared multi-agent catalog.
-
Handle large Codex reviews more reliably with better prompt budgeting and fallback behavior.
-
Use
prekfor pre-commit auto-fixes and lint hook management. -
Honor
NO_COLORconsistently on TUI detail screens. -
Keep
roborev refinebranch reviews on the configured agent. -
Run worktree-based reviews and hooks in the correct repository context.
-
Fix Copilot permission failures in non-interactive runs.
v0.48.0
Improvements
- Review agents run in a read-only sandbox to better protect your checked-out worktree.
- Review jobs avoid
.git/index.lockcontention, improving reliability when Git is active.
Bug Fixes
- Development builds no longer show update notifications.
v0.47.2
Bug Fixes
roborev fixcorrectly honors--all-branchesand--branchwhen run from Git worktrees.