Skip to content
roborev
GitHub stars

Changelog

All notable changes to roborev, grouped by minor release.

0.51

2026-04-08

New features

  • OpenAPI spec for the daemon REST API, served at /openapi.json (OpenAPI 3.1.0 via Huma). The spec covers the primary query and mutation endpoints used by integrations (jobs, reviews, comments, repos, branches, status, summary, cancel, rerun, close). Internal endpoints used by the CLI, TUI, and daemon subsystems (enqueue, streaming, sync, fix orchestration) are not part of the OpenAPI surface. See Streaming & Daemon API.
  • Cascading review_min_severity setting and --min-severity flag on roborev review to filter review findings by severity. The setting cascades from CLI flag to per-repo .roborev.toml to global config.toml, matching the existing pattern for fix_min_severity and refine_min_severity. Global defaults for all three (review_min_severity, fix_min_severity, refine_min_severity) are now supported in global config. See Configuration.

Improvements

  • Branch review prompts include per-commit review context. When reviewing a commit range, the prompt includes summaries and verdicts from individual per-commit reviews, with instructions to focus on cross-commit interactions instead of re-raising known issues.
  • Fix prompts include user comments and prior tool attempts. Developer comments and previous automated fix attempts are separated and included in the prompt, giving the fix agent more context about what has already been tried and what the developer flagged.
  • Global reasoning defaults are honored consistently across review, fix, refine, and related workflows. Resolution order: explicit CLI flag > per-repo config > global config > default.
  • The TUI lets you inspect the full prompt while a job is still queued, before it starts running. Press p on any queued job that has a stored prompt (task, fix, compact, insights).

Bug fixes

  • roborev fix correctly discovers open jobs on the current branch. Previously, it could include jobs from unrelated branches or miss jobs when run from main due to unreachable SHAs from squashed or amended commits.
  • Codex sandbox compatibility improvements. A new disable_codex_sandbox config option bypasses bwrap sandboxing on systems where it is unavailable. Read-only sandboxed reviews fall back to inline diff snapshots when .git/ is inaccessible to the agent. See Configuration.
  • Codex review jobs now store and display the actual command line used, fixing incorrect command reporting in the TUI.
  • CI repo matching resolves ambiguous repositories (multiple repos sharing the same git identity) by preferring auto-cloned repos instead of failing.
  • GitHub Actions release checksums use the expected SHA256SUMS filename.

0.50

2026-03-31

New features

  • auto_close_passing_reviews config option to automatically close reviews that pass with no findings. When enabled, pass reviews are closed immediately instead of staying open in the queue. See Configuration.
  • Bundled roborev-refine skills for Claude Code and Codex to run iterative review-fix-review loops from within an agent session. The skill performs the full refine workflow inline (review, fix, commit, re-review) rather than shelling out to the CLI. See Agent Skills.
  • Bundled systemd service and socket unit files for Linux daemon deployments. The service uses Type=notify for readiness signaling. Socket activation is supported for on-demand daemon startup. See Persistent Daemon.

Improvements

  • The TUI updates instantly by subscribing to the daemon event stream (SSE) instead of polling on a timer. Polling is retained as a 15-second fallback.
  • CI review prompts now include human PR discussion (issue comments, review summaries, and inline review comments) from trusted collaborators with maintain or admin access. Discussion is treated as untrusted context with safety guardrails.
  • The daemon socket path prefers $XDG_RUNTIME_DIR/roborev/daemon.sock when the variable is set and points to an existing absolute directory. Falls back to the platform temp directory otherwise. See Configuration.

Bug fixes

  • Preserve the requested model when rerunning reviews. Previously, rerunning a review could resolve a different model from config defaults instead of preserving the model specified in the original request. A separate requested_model field now tracks explicit user intent.
  • Enforce a batch_timeout (default: 3 minutes) on CI PR comment batches to prevent indefinite hangs when some jobs in a multi-agent batch get stuck. When the timeout expires, available results are posted and remaining jobs are canceled. See CI Options Reference.

0.49

2026-03-24

New features

  • roborev insights command analyzes failing code reviews to identify recurring patterns, hotspot areas, noise candidates, and guideline gaps. Outputs actionable suggestions for improving review_guidelines in .roborev.toml. Runs as a daemon-backed job, queued and tracked like reviews. See Commands.
  • Unix domain socket support for CLI-to-daemon communication on Unix systems. Set server_addr = "unix://" in ~/.roborev/config.toml to listen on /tmp/roborev-{UID}/daemon.sock instead of TCP loopback. Socket permissions (0600) enforce per-user access control. See Configuration.
  • ROBOREV_COLOR_MODE environment variable to force auto, dark, light, or none color output across all TUI screens and CLI rendering. See Configuration.

Improvements

  • Skill installation and status reporting use a shared multi-agent catalog. roborev skills now shows per-agent status (installed, outdated, not installed, no agent) for both Claude Code and Codex. Adding future agents requires a single catalog entry.
  • Large Codex reviews are more reliable. Prompt budgeting is now configurable via max_prompt_size (per repo) and default_max_prompt_size (global), with smart fallback instructions that guide Codex to read diffs locally when they exceed the prompt budget. Diffs are read in bounded chunks with UTF-8-safe truncation.
  • Pre-commit auto-fixes and lint hook management now use prek instead of a custom shell script (roborev development workflow only).

Bug fixes

  • NO_COLOR is honored on TUI review and prompt detail screens. Previously, glamour markdown rendering defaulted to TrueColor regardless of NO_COLOR.
  • roborev refine branch reviews now use the configured review agent instead of the fix agent.
  • Reviews and hooks for commits made in git worktrees now run in the correct worktree directory. A worktree_path field is persisted per job so agents and hooks operate on the right branch.
  • Copilot reviews no longer fail with permission denials in non-interactive (daemon) mode. The agent now uses --allow-all-tools with a deny-list for destructive operations in review mode.

0.48

2026-03-18

Improvements

  • Review agents now run in a read-only sandbox. Codex review jobs use --sandbox read-only instead of --full-auto, matching Claude Code’s existing read-only tool restrictions. Agentic mode (fix, refine, --agentic) is unchanged. All agent subprocesses set GIT_OPTIONAL_LOCKS=0 to avoid contending with the user’s own git operations.
  • --open and --unaddressed flags on roborev fix are deprecated. Open job discovery is now the default behavior when no positional job IDs are provided. The flags are hidden and silently ignored for backwards compatibility.
  • --branch <name> flag added to roborev fix for cross-branch fixing without switching branches.
  • Skip update notifications in development builds.

Bug fixes

  • Avoid .git/index.lock contention during reviews by setting GIT_OPTIONAL_LOCKS=0 in agent subprocess environments, reducing conflicts with concurrent git operations.
  • Fix --all-branches and --branch filtering when running roborev fix from a git worktree. The branch override was not being threaded through to filterReachableJobs, causing it to filter by the worktree’s branch instead of the requested branch.

0.47

2026-03-17

New features

  • roborev summary command shows aggregate review statistics: pass/fail trends, per-agent effectiveness, review duration percentiles, fix resolution rates, and per-repo breakdowns. Scopes to the current repo by default; use --all for cross-repo summary. Supports --since, --branch, --repo, and --json flags. See Commands.
  • TUI control socket for programmatic interaction with running TUI instances. External tools can query state and trigger mutations (filter, select, close, cancel, rerun, quit) over a Unix domain socket using a newline-delimited JSON protocol. Runtime metadata is written to ~/.roborev/tui.{PID}.json for discoverability. See TUI Control Socket.
  • --no-quit flag on roborev tui suppresses keyboard quit (q) in queue and tasks views, allowing external controllers to manage the TUI lifecycle. The quit control command still works regardless.
  • Token usage tracking: agent token consumption (peak context tokens and total output tokens) is automatically recorded after each job completes and displayed in the TUI review header and roborev show output. Requires agentsview to be installed. roborev backfill-tokens retroactively fetches token data for completed jobs that have session IDs but no stored usage.
  • opencode_cmd config key to override the OpenCode executable path, matching the existing pattern for other *_cmd overrides.

Improvements

  • Common lockfiles and generated files (package-lock.json, yarn.lock, go.sum, Cargo.lock, uv.lock, and others) are excluded from review diffs by default. Add custom patterns via exclude_patterns in global or per-repo config. Security reviews skip repo-level exclude patterns to prevent suppression of sensitive files. See Configuration.
  • maximum reasoning level (aliases: max, xhigh) maps to Codex’s xhigh reasoning effort. For agents without an xhigh equivalent, it maps to thorough. See Reasoning Levels.
  • Session ID column in the TUI queue view.
  • Column checkboxes in the TUI options menu respond to mouse clicks.
  • Long comment text word-wraps in the TUI review pane.
  • The TUI elapsed timer updates every second instead of only on data refreshes.
  • Skill names switched from colon syntax (roborev:fix) to hyphenated syntax (roborev-fix) for compatibility with GitHub Copilot CLI. Run roborev skills update to apply the new names. Both Claude Code and Codex skills are updated.

Bug fixes

  • Fix GetCurrentBranch returning a heads/-prefixed branch name when git refs are ambiguous.
  • Update Gemini defaults and fall back cleanly when the configured model is unavailable.

0.46

2026-03-11

Improvements

  • Agent availability checks now honor *_cmd config overrides (claude_code_cmd, codex_cmd, cursor_cmd, pi_cmd). Previously, custom agent commands were ignored during availability detection, so an agent could appear unavailable even when the configured binary was in PATH. See Agent Command Overrides.
  • The TUI review screen now displays the branch name stored with the review instead of resolving it dynamically via git name-rev. In worktree setups where the same SHA is reachable from multiple refs, the old behavior could display the wrong branch.

Bug fixes

  • Fix the post-commit hook sending the worktree path instead of the main repository root to the daemon when running inside a linked git worktree. This caused commits to be registered under a phantom repo entry.
  • Fix roborev fix --open, --list, and --batch discovering reviews from other worktrees. Jobs are now filtered to only those reachable from the current worktree’s HEAD or matching its branch name.
  • Fix the post-commit hook not firing in linked git worktrees when core.hooksPath is set to a relative path. Relative paths are now resolved against the main repository root instead of the worktree root. init, install-hook, and uninstall-hook also normalize the hooks path and fail early if it cannot be resolved.
  • Add JSONL post-commit hook logging to ~/.roborev/post-commit.log so that silent hook failures leave an audit trail with timestamps, repo paths, and failure reasons. See Troubleshooting.

0.45

2026-03-08

New features

  • --min-severity flag for roborev fix and roborev refine to limit fixes to findings at or above a chosen severity (low, medium, high, critical). Also configurable per repo via fix_min_severity and refine_min_severity in .roborev.toml. When all findings in a review fall below the threshold, refine automatically closes the review instead of treating it as a fix failure.
  • Experimental: reuse_review_session config option (global and per repo) to resume prior agent sessions on the same branch, reducing token usage and review latency on active branches. See Session Reuse.

Improvements

  • roborev show now displays comments after the review output, matching the TUI review detail view.
  • Copied reviews (TUI y key) now include review comments, giving the fix agent more context when you paste into an agent session.
  • Agent tool-call narration (text the agent emits before tool calls) is stripped from persisted review output across all agents.
  • Daemon status details are hidden from the review detail view in the TUI; the queue view is unchanged.
  • Review prompts now instruct agents not to build projects, run tests, or execute code during review.
  • roborev config set and roborev init now produce commented TOML output with inline descriptions for each field.

Bug fixes

  • Fix roborev compact using the wrong branch inside git worktrees. It was resolving the main checkout’s branch instead of the worktree’s branch.
  • Fix workflow model fallback so it uses the selected agent’s actual default model instead of the global default.
  • Job log files are no longer permanently lost when the initial file open fails under resource pressure. The new log writer retries and buffers output until disk logging recovers.
  • Timed-out review jobs now unwind reliably instead of appearing to run past the configured job_timeout. Timeout errors are recorded as agent timeout after <duration> for clearer reporting in the TUI and hooks.

0.44

2026-03-07

New features

  • mouse_enabled global config flag (and TUI options menu toggle) to disable mouse interactions in the TUI.
  • roborev post-commit command and post_commit_review repo config to control post-commit review behavior, including branch review workflows.
  • Webhook review hooks (type = "webhook") for external integrations.
  • excluded_commit_patterns repo config for skipping reviews based on commit message substrings.
  • auto_filter_branch global config to automatically filter the TUI to the current branch or worktree on startup.

Improvements

  • Tighter review prompts and more consistent verdict parsing.
  • Daemon/client mismatch warnings surfaced through daemon status output.

Bug fixes

  • Retry fix daemon calls automatically after a daemon restart.
  • Fix daemon startup restart loops.
  • Fix Enter key handling in the TUI inside embedded terminals.
  • Fix cases where the TUI fails to close cleanly.

0.43

2026-03-06

New features

  • default_backup_model config option to control the fallback model used by agent workflows when the primary model is unavailable.
  • advanced.enable_tasks config flag to opt in to the TUI background tasks workflow (fix jobs, patch application, and rebasing). This workflow was previously enabled by default and has been moved behind a flag to avoid confusion about the primary review workflow.

Improvements

  • Ctrl-D quits the TUI as an additional shortcut alongside q.
  • Improved built-in agent skill definitions for more reliable matching, and expanded agent configuration documentation.

Bug fixes

  • Agent resolution for review, analyze, fix, and refine commands now selects the intended agent more reliably.
  • CLI --agent overrides no longer inherit the wrong default_model from configuration.

0.42

2026-03-05

New features

  • Multi-repo workspace support: roborev list looks in immediate child subfolders for repos, and roborev review suggests repo-level review commands.
  • Cursor agent support.
  • Pi coding agent support.
  • Save generated patch files to disk from the TUI Tasks view.

Improvements

  • Skip review throttling when a new push supersedes an in-progress review. The old review is canceled and the new one starts immediately.
  • Validate configured agent names and reject unknown agents earlier.

Bug fixes

  • Improve Claude review failure reporting so agent errors are captured and surfaced correctly.

0.41

2026-03-04

Bug fixes

  • Restore a separate P/F verdict column in the TUI queue so review outcomes are easier to scan.

0.40

2026-03-03

New features

  • ACP (Agent Client Protocol) support: run reviews through any ACP-compatible agent via the [acp] config section.
  • Kiro agent integration via kiro-cli.
  • Configurable PR comment upsert: update existing roborev PR comments instead of posting duplicates (ci.upsert_comments).

Improvements

  • Renamed review status terminology to closed/open across CLI, TUI, and API. roborev close and roborev fix --open replace the legacy command/flag aliases (which are still accepted).
  • Combined the separate Status and P/F columns in the TUI queue into a single Status column with color-coded states (Queued, Running, Pass, Fail, Error, Canceled).
  • Column customization in the TUI: press o to reorder or toggle column visibility. New column_borders, column_order, and task_column_order config options.
  • Mouse copy/paste in TUI content views; long stderr lines wrap in log views.
  • Visual polish across TUI queue, review, and task screens: tighter column spacing, box-drawing separators, right-aligned elapsed column.
  • Deprecated the /roborev:address skill in favor of /roborev:fix.

Bug fixes

  • Fixed UTF-8 truncation when composing PR comments.
  • Fixed command/footer parsing by trimming trailing blank lines and enforcing -- separators.

0.39

2026-02-28

New features

  • Compact mode for better usability on short terminal windows.
  • Distraction-free toggle in the TUI for a cleaner review experience.

Bug fixes

  • Custom fix instructions now include full review context during fix generation.

0.38

2026-02-26

New features

  • Kilo agent support via the kilo CLI.
  • roborev wait accepts multiple job IDs in a single command.

Improvements

  • TUI task view supports mouse interactions (click to select, double-click to view).
  • roborev update manages the daemon lifecycle for smoother upgrades.

Bug fixes

  • Use ANTHROPIC_API_KEY for the OpenCode agent in GitHub Actions workflows.
  • roborev fix skips reviews that already have a PASS verdict.

0.37

2026-02-25

Improvements

  • TUI help bar restyled with two-tone key hints and aligned columns for easier shortcut scanning.
  • Unified stream output formatting across CLI and TUI views for more consistent display.

Bug fixes

  • Show the correct roborev version when installed via go install.

0.36

2026-02-24

New features

  • roborev tui --repo and --branch flags to launch the TUI pre-filtered to a specific repository or branch. Without a value, each flag resolves to the current repo/branch. With = syntax (e.g. --repo=/path/to/repo, --branch=feature-x), the value is used directly. When set via flags, the filter is locked and cannot be changed in the TUI.
  • Inline fix panel in the TUI review view: press F while viewing a review to open a fix prompt at the bottom of the screen instead of a full-screen modal. Tab toggles focus between the review content and the fix input. Enter submits, Esc cancels.
  • Shell completions for --agent and --reasoning flags across all commands that accept them (init, review, run, fix, analyze, refine).
  • OpenCode JSON stream support: the OpenCode agent now uses --format json for structured JSONL output, integrated into the unified stream formatter for consistent progress rendering.
  • CI repository matching with wildcard patterns and exclusion lists. ci.repos entries now support glob patterns (e.g. "myorg/*", "myorg/api-*") using path.Match syntax. New exclude_repos field filters out matching repos, and max_repos (default: 100) caps the total expanded count. Wildcard results are cached for one hour.

Improvements

  • TUI help bar uses table-based rendering for consistent column alignment across all views.

Bug fixes

  • --all-branches now implies --open on roborev fix and roborev refine, removing the need to pass both flags.
  • Patch application in git worktrees resolves the correct worktree path via git worktree list, fixing failures when the branch is checked out in a non-default worktree location.
  • Temporary command execution uses explicit file sync and retry with exponential backoff to prevent intermittent text file busy (ETXTBSY) races on Linux.

0.35

2026-02-23

New features

  • Shell completion for roborev analyze command types: tab-complete analysis type names (e.g. roborev analyze <TAB> suggests refactor, complexity, etc.).
  • Persistent job logs: agent output is written to ~/.roborev/logs/jobs/ as NDJSON so review activity survives daemon restarts.
  • Unified log viewer: roborev log <job-id> renders stored job output on the CLI, and pressing l in the TUI opens a scrollable log viewer with live polling for running jobs. roborev log clean removes old log files.

Improvements

  • Test and production runtime data are isolated so go test runs do not pollute ~/.roborev/ logs or interfere with the production daemon.
  • CLI and TUI streaming output uses gutter-grouped tool calls, markdown text wrapping, and Codex reasoning item rendering for clearer review progress.

Bug fixes

  • Handle empty Git refs when fixing compact review jobs to prevent fix-flow failures. The server resolves a usable ref from the parent job’s branch or falls back to HEAD, and the TUI shows a confirmation modal when no ref is available.

0.34

2026-02-22

New features

  • roborev ci review: daemon-free batch reviews for CI pipelines with auto-detection of GitHub Actions environment variables (GITHUB_REPOSITORY, GITHUB_REF, GITHUB_EVENT_PATH).
  • roborev init gh-action: generates a GitHub Actions workflow file with SHA256-verified roborev installation and agent setup.
  • TUI fix jobs: press F on a completed review to launch a background fix in an isolated worktree. New Tasks view (T) for managing fix jobs and applying patches.
  • CI poller auto-clone: repos in ci.repos no longer require a local roborev init checkout. The poller clones them automatically to ~/.roborev/clones/.
  • Quota-aware agent cooldown: agents that hit hard quota limits enter a timed cooldown (default 30 min) with automatic failover to backup agents. CI comments show “skipped (quota)” instead of “failed”.
  • Daemon activity logging for better operational visibility.

Improvements

  • Review verdicts are stored for reuse in later review workflows.

Bug fixes

  • Fix jobs now create worktrees at the reviewed commit instead of HEAD, preventing patches against the wrong revision.
  • Database migration no longer crashes on databases with quoted table names from prior ALTER TABLE migrations.
  • Missing git origin remote treated as confirmed mismatch for auto-clone instead of a transient error.
  • Fixed a data race between WorkerPool.Start and WorkerPool.Stop.

0.33

2026-02-17

New features

  • roborev compact command to verify and consolidate open review findings, reducing false positives and merging related findings from multiple reviews into a single consolidated review.
  • Backup-agent failover: automatically retry failed jobs with a secondary agent when the primary fails (e.g. fall back to Claude Code when Codex rate-limits).
  • GitHub commit status checks: the CI poller posts pending/success/failure statuses on PR commits when GitHub App auth is configured.
  • Ref-aware configuration: the CI poller reads .roborev.toml from the PR branch’s git ref, so configuration can vary by branch.
  • --label flag on roborev run for custom labels displayed in the TUI.

Improvements

  • Consolidated review guidelines for more consistent review output across commands.
  • Hardened CI and hook workflows for more reliable automated runs.

Bug fixes

  • Post-rewrite hook preserves review history across rebases by remapping commit SHAs when patch content is unchanged.
  • Skip hook upgrade checks in CI mode to avoid CI interruptions.

0.32

2026-02-16

New features

  • roborev wait command to block until a review job completes, improving scripting and CI flows.
  • Refine targeting flags so you can run roborev refine against specific findings.
  • Unified TUI tree filter with lazy branch loading, search, and current-directory prioritization.

Improvements

  • Improved TUI hint bar to make available actions clearer.
  • Removed the hardcoded OpenCode model so model selection follows your configuration.

Bug fixes

  • Fixed TUI Cursor cancel behavior and corrected closed/open stats display.
  • Fixed agent prompt handling on Windows to avoid the 32KB command-line limit.
  • Fixed refine loops so git hook failures no longer break execution.
  • Stripped CLAUDECODE when spawning the claude-code agent to prevent environment leakage.

0.31

2026-02-11

New features

  • roborev config subcommands (get, set, list) for viewing and managing configuration from the CLI.
  • --branch <name> flag on roborev analyze and explicit branch names in roborev review --branch.

Improvements

  • Refreshed built-in Claude and Codex skill guides for review/refine/respond/fix workflows.

0.30

2026-02-11

New features

  • TUI renders Markdown in review output for clearer formatting.

Improvements

  • TUI output is sanitized and escaped to prevent control sequences from breaking terminal rendering.

0.29

2026-02-10

New features

  • review and review-branch skills for Codex and Claude to run code reviews from agent skills.
  • design-review-branch skills for Codex and Claude.

Improvements

  • Normalized skill invocation patterns for more consistent matching.
  • Improved Codex stream handling with stronger merge guarding.

Bug fixes

  • Fixed cases where the Codex agent produces no visible CLI output.
  • Fixed range reviews that fail when the start point is the repository root commit.

0.28

2026-02-10

New features

  • Server-side filtering for review/job lists.
  • Automatic TUI filtering to narrow visible reviews and jobs.
  • Filter metrics to show what filtering matches.

Improvements

  • Clearer TUI command display and improved prompt navigation.

Bug fixes

  • Prevented the daemon from inheriting GIT_DIR from Git hook environments.

0.27

2026-02-09

New features

  • --type flag for design and security reviews from the CLI.
  • Jump-to-top shortcut (g) in the TUI.
  • Built-in design review skill templates for Codex and Claude.

Bug fixes

  • Fixed review-type consistency so selected modes are applied reliably across commands.

0.26

2026-02-08

New features

  • CI poller that detects GitHub pull requests and queues reviews automatically.
  • GitHub App integration for authenticated PR review workflows.
  • Persistent CI review tracking that survives daemon restarts.
  • hide_closed_by_default config option for the TUI.

Improvements

  • Expanded configuration to support CI polling and GitHub App settings.
  • Default Gemini model set so Gemini works without explicit model configuration.
  • Hardened agent integrations for improved reliability.

0.25

2026-02-04

New features

  • roborev list command for viewing stored reviews.
  • --json flag on roborev show for machine-readable output.
  • Color-coded Closed column in the TUI.

Improvements

  • TUI queue view displays JobID instead of ID for clearer identification.

Bug fixes

  • Fixed verdict detection for Severity: Level format.
  • Fixed hook v1 to v2 upgrade by stripping & and documenting install-hook --force.

0.24

2026-02-03

Improvements

  • Show available fixes in the roborev fix list.
  • Fail fast when task jobs are missing a prompt.

Bug fixes

  • Prevented wrong agent selection and duplicate reviews from the post-commit hook.

0.23

2026-02-02

New features

  • /roborev:fix skill to address multiple review findings in one pass.
  • {findings} template variable for hook commands.

Improvements

  • Show skill status in roborev skills output.
  • Upgrade post-commit hook on init to keep tooling up to date.

Bug fixes

  • Fixed post-commit hook backgrounding to avoid blocking or hangups.

0.22

2026-01-31

New features

  • Review hooks system to run shell commands when reviews complete or fail.
  • --batch flag on roborev fix for batch operation.

Improvements

  • Rewritten README documenting the coding agent workflow.

Bug fixes

  • Fixed hook tests for portability across environments.

0.21

2026-01-30

New features

  • Cursor agent support.
  • check-agents command to list and smoke-test available agents.
  • --open flag on fix for batch fixing.
  • show --prompt to display the prompt sent to the agent.

Improvements

  • Improved daemon resilience and overall UX.
  • Include current UTC date in review prompts for temporal context.

Bug fixes

  • Fixed shell wildcard expansion in analyze when run from subdirectories.
  • Prevented duplicate review jobs when enqueueing.
  • Fixed branchless jobs not included when running fix.

0.20

2026-01-29

New features

  • roborev analyze for built-in code analysis workflows.
  • roborev fix to apply guided fixes from analysis results.

Bug fixes

  • Fixed cosmetic issues in repo stats display.
  • Fixed zero “Created” date in roborev repo show.

0.19

2026-01-27

New features

  • Workflow-specific configuration keys and --fast shorthand flag.
  • Branch column in the TUI with filtering support.
  • --local flag to run reviews without starting the daemon.

Improvements

  • Improved TUI row selection styling.

Bug fixes

  • Fixed branch filter returning no results when fetch is limited.
  • Fixed false negative verdicts when severity labels are present.
  • Fixed make install to avoid using go install.

0.18

2026-01-26

New features

  • tail command to view streaming agent output.
  • Support for multiple clones running concurrently.
  • Automatic terminal color adaptation for light/dark themes.

Improvements

  • Show model name and reorganized Review screen layout.

Bug fixes

  • Fixed address API and CLI to use job_id correctly.

0.17

2026-01-25

New features

  • Configurable model selection for all agents.
  • Gemini-specific preamble support for run tasks.
  • TUI commit viewer, help modal, and clearer navigation feedback.

0.16

2026-01-24

New features

  • Layered Escape key behavior to clear filters one level at a time.
  • Gemini-specific review template with upfront summary requirement.

Improvements

  • Renamed prompt command to run for clearer CLI usage.
  • Improved daemon lifecycle management for safer start/stop.

Bug fixes

  • Fixed TUI flickering when the queue is empty with filters applied.
  • Fixed edge cases in daemon shutdown.

0.15

2026-01-23

New features

  • Config hot-reload for the daemon.
  • Factory Droid agent support.
  • y hotkey to copy review content to the clipboard.
  • Review metadata header in clipboard yank content.
  • PowerShell installer and ARM64 builds for Windows.

Improvements

  • Flash notifications for incomplete jobs in the TUI.
  • Homebrew tap integration for easier installation.

Bug fixes

  • Fixed multi-byte character handling in TUI text input.
  • Fixed Codex agent stdin handling on Windows.

0.14

2026-01-21

New features

  • TUI respond modal to capture review responses and include them in future prompts.

Bug fixes

  • Fixed TUI rendering artifacts when scrolling with page up/down.

0.13

2026-01-20

New features

  • PostgreSQL sync to share reviews across multiple machines.

Improvements

  • Simplified install.sh and moved docs screenshots to roborev-docs repo.
  • Instruct reviewers to skip commit message review.

Bug fixes

  • Fixed race condition that caused closed items to briefly reappear.
  • Fixed markdown formatting in verdict parsing.
  • Fixed sync now to connect automatically when the daemon is not yet connected.

0.12

2026-01-19

New features

  • --since option on roborev review to scope reviews to recent changes.
  • Gemini support for roborev refine.
  • Copilot and OpenCode agent support.

Improvements

  • Default allow_unsafe_agents to true for refine when using Claude.
  • Improved TUI rendering and presentation.

Bug fixes

  • Fixed TUI rendering glitches and layout issues.

0.11

2026-01-18

New features

  • roborev prompt command for custom agent tasks.
  • roborev repo command for managing tracked repositories.
  • Nix flake app entry for roborev.

Improvements

  • Claude Code compatibility for roborev refine.
  • Expanded daemon API to support repo and prompt operations.

0.10

2026-01-16

New features

  • roborev skills install command to install bundled agent skills.
  • Bundled skills for Claude Code and Codex (address/respond workflows).

0.9

2026-01-14

New features

  • refine command for automated review fixing.

Improvements

  • Allow roborev refine on main with --since, waiting for in-progress reviews.
  • Use configured display_name in the filter modal.

Bug fixes

  • Fixed queue cursor behavior when hide-closed is active and closing from the review screen.

0.8

2026-01-13

New features

  • Renamed enqueue to review with a cleaner CLI interface.
  • --dirty flag to review uncommitted changes.
  • --wait flag to keep the CLI open until review completes.

0.7

2026-01-11

New features

  • r hotkey to rerun failed/canceled jobs or start a new review.
  • roborev stream command for JSONL event streaming.
  • excluded_branches and display_name config options.
  • Nix flake for building and development.

Improvements

  • Full commit message bodies included in review prompts.
  • Clearer new release notifications.

Bug fixes

  • Fixed git worktrees being treated as separate repositories.
  • Fixed false positive “failed” reviews.

0.6

2026-01-10

New features

  • h hotkey to hide closed reviews.
  • Branch display in the TUI review view.
  • Distinct [CLOSED] color styling in review view.

Improvements

  • Improved verdict parsing.
  • Improved TUI height sizing and review ID display.

Bug fixes

  • Fixed TUI height sizing display issues.

0.5

2026-01-09

New features

  • Filter-by-repo modal in the TUI.
  • ROBOREV_DATA_DIR env var to override the data directory.
  • Configurable job timeout.
  • TUI pagination for large review lists.
  • Keyboard navigation between reviews without returning to the list.
  • P/F (Pass/Fail) verdict column in the TUI queue.

Improvements

  • TUI views fit terminal width dynamically.
  • More robust executable path handling for hooks.

0.4

2026-01-08

New features

  • roborev update command to check for and install updates.
  • TUI notification when a new version is available.
  • Husky git hook manager support.

Improvements

  • Automatic .git/hooks directory creation.
  • Respect core.hooksPath for git operations.
  • Refactored post-commit hook for improved security and silent operation.
  • Detect rebase state and skip reviews during rebase.

Bug fixes

  • Fixed version comparison for dev builds.
  • Fixed Windows path detection for hook locations.

0.3

2026-01-07

New features

  • Job cancellation with x key in the TUI (terminates agent subprocess).
  • uninstall-hook command.

Bug fixes

  • Fixed TUI selection highlight to cover the full line.
  • Fixed job cancellation persistence and race conditions.
  • Fixed migration handling for foreign keys and ALTER TABLE ordering.

0.2

2026-01-06

New features

  • Project-specific review guidelines in .roborev.toml.
  • Closed status tracking with a dedicated closed column and toggle.
  • Prompt inspection in the TUI.
  • Page up/down navigation in the TUI.
  • Daemon version tracking with auto-restart on upgrade.
  • Gemini CLI and Copilot CLI agent support.
  • OpenCode agent support.
  • Automatic retry for failed reviews (up to 3 attempts).

Improvements

  • Optimistic updates for close toggle.
  • Compact timestamp format in the TUI queue.
  • Daemon version displayed in the TUI and CLI.

Bug fixes

  • Fixed TUI queue edge cases for empty queues and navigation.
  • Fixed daemon stop behavior and restart reliability.
  • Fixed SQLite datetime parsing for TUI timestamps.
  • Fixed retry job atomicity.

0.1

2026-01-05

Initial release.

  • Pure-Go SQLite driver for static binaries.
  • --addr normalization to add http:// prefix if missing.