project devlogs / roborev
roborev Catch bugs while your agents work, with local background reviews of every commit and a persistent queue of findings to resolve
GitHub ↗ Website ↗
By week By project Choose project roborev ↓ AgentsView msgvault roborev kata Kenn Forge docbank Ghosthub kwt BenchDB roborev source code growth Source code measured at 36 snapshots, from 3,807 lines on 2026-01-05 to 263,517 lines on 2026-09-05. Full measurements are in the table below. 0 133,071 266,142 2026-01-05: 3,807 source lines 2026-01-11: 13,108 source lines 2026-01-18: 26,687 source lines 2026-01-25: 42,195 source lines 2026-02-01: 50,293 source lines 2026-02-08: 51,632 source lines 2026-02-15: 63,910 source lines 2026-02-22: 79,899 source lines 2026-03-01: 94,230 source lines 2026-03-08: 105,852 source lines 2026-03-15: 104,759 source lines 2026-03-22: 111,465 source lines 2026-03-29: 115,403 source lines 2026-04-05: 120,889 source lines 2026-04-12: 122,902 source lines 2026-04-19: 125,757 source lines 2026-04-26: 131,724 source lines 2026-05-03: 133,151 source lines 2026-05-10: 135,720 source lines 2026-05-17: 136,889 source lines 2026-05-24: 138,392 source lines 2026-05-31: 140,086 source lines 2026-06-07: 156,558 source lines 2026-06-14: 162,613 source lines 2026-06-21: 166,215 source lines 2026-06-28: 172,469 source lines 2026-07-05: 178,894 source lines 2026-07-12: 182,049 source lines 2026-07-19: 185,878 source lines 2026-07-26: 186,788 source lines 2026-08-02: 189,459 source lines 2026-08-09: 200,895 source lines 2026-08-16: 243,090 source lines 2026-08-23: 254,172 source lines 2026-08-30: 266,142 source lines 2026-09-05: 263,517 source lines 2026-01-05 2026-09-05 roborev source code growth Source code measured at 36 snapshots, from 3,807 lines on 2026-01-05 to 263,517 lines on 2026-09-05. Full measurements are in the table below. 0 133.1K 266.1K 2026-01-05: 3,807 source lines 2026-01-11: 13,108 source lines 2026-01-18: 26,687 source lines 2026-01-25: 42,195 source lines 2026-02-01: 50,293 source lines 2026-02-08: 51,632 source lines 2026-02-15: 63,910 source lines 2026-02-22: 79,899 source lines 2026-03-01: 94,230 source lines 2026-03-08: 105,852 source lines 2026-03-15: 104,759 source lines 2026-03-22: 111,465 source lines 2026-03-29: 115,403 source lines 2026-04-05: 120,889 source lines 2026-04-12: 122,902 source lines 2026-04-19: 125,757 source lines 2026-04-26: 131,724 source lines 2026-05-03: 133,151 source lines 2026-05-10: 135,720 source lines 2026-05-17: 136,889 source lines 2026-05-24: 138,392 source lines 2026-05-31: 140,086 source lines 2026-06-07: 156,558 source lines 2026-06-14: 162,613 source lines 2026-06-21: 166,215 source lines 2026-06-28: 172,469 source lines 2026-07-05: 178,894 source lines 2026-07-12: 182,049 source lines 2026-07-19: 185,878 source lines 2026-07-26: 186,788 source lines 2026-08-02: 189,459 source lines 2026-08-09: 200,895 source lines 2026-08-16: 243,090 source lines 2026-08-23: 254,172 source lines 2026-08-30: 266,142 source lines 2026-09-05: 263,517 source lines 2026-01-05 2026-09-05 +259,710 lines since 2026-01-05
Measurements & method SCC code lines (nonblank, noncomment) in tracked source files, including tests, SQL, HTML, and CSS. Excludes prose, shell scripts, data/config files, vendored dependencies, build output, generated and minified files. Counts the checked-out tree, not accumulated additions.
Snapshots follow the repository’s default branch. Lines measure codebase size, not work completed. Historical counts can fall when code is removed.
Measured with scc 3.7.0. Updated 2026-09-05.
Source line measurements for roborev Date Source lines Files Commit 2026-09-05 263,517 899 a19babb 2026-08-30 266,142 881 77e7452 2026-08-23 254,172 863 14c51f8 2026-08-16 243,090 824 bafa98f 2026-08-09 200,895 673 37a5a36 2026-08-02 189,459 634 1164a3e 2026-07-26 186,788 631 8cf0919 2026-07-19 185,878 627 b5146c2 2026-07-12 182,049 617 b97a3d8 2026-07-05 178,894 605 46b58b9 2026-06-28 172,469 587 0560c8a 2026-06-21 166,215 571 d7f8eea 2026-06-14 162,613 548 473c539 2026-06-07 156,558 503 eb6b8b5 2026-05-31 140,086 410 d7400d1 2026-05-24 138,392 408 875b0a4 2026-05-17 136,889 408 3172d3b 2026-05-10 135,720 405 d9aae1f 2026-05-03 133,151 395 51f84e1 2026-04-26 131,724 393 4879fa8 2026-04-19 125,757 367 d9aafdf 2026-04-12 122,902 363 dc18257 2026-04-05 120,889 353 dc42e2a 2026-03-29 115,403 346 ea575f3 2026-03-22 111,465 321 1ae1f8c 2026-03-15 104,759 295 ff3a202 2026-03-08 105,852 292 85f4787 2026-03-01 94,230 272 e84fb1e 2026-02-22 79,899 172 6a81db1 2026-02-15 63,910 131 9d1eeee 2026-02-08 51,632 106 f53d70b 2026-02-01 50,293 105 82e4a8c 2026-01-25 42,195 84 f8f6658 2026-01-18 26,687 60 bf49003 2026-01-11 13,108 37 a51d87e 2026-01-05 3,807 29 cdd5f65
Project timeline Newest first 712 merged PRs · 115 releases · 34 weeks in the devlog
2026-08-24 to 2026-08-30
12 PRs merged (12 Kenn, 0 community) Weekly diff: +15,091 / −2,155 lines
Define custom review types to tailor reviews to different needs. Use branch-specific review configuration in experiments. Batch automatic post-commit reviews instead of running each one separately. See why a review failed in the web UI, which also loads faster on first open. Use roborev list --all-branches to include reviews outside the current branch, and skip CI reviews for pull requests with configured labels. 2026-08-17 to 2026-08-23
21 PRs merged (19 Kenn, 2 community) Weekly diff: +17,790 / −4,838 lines
Self-updates now coordinate with active reviews instead of disrupting work in progress. The browser UI now supports base paths and authentication from trusted proxies. It also explains when missing assets make the UI unavailable. Jobs now fail promptly after agent process errors, delayed prices recover automatically, and empty reviews no longer post CI errors. Antigravity reviews now handle large prompts and can use the pwd command. Native selection works again in split review details, so you can copy individual findings. 2026-08-10 to 2026-08-16
17 PRs merged (16 Kenn, 1 community) Weekly diff: +58,363 / −1,960 lines
Use Roborev through its new native browser application. Review changes in a split-screen view in the terminal interface. Restart the daemon without interrupting reviews that are still running. See active snooze status in both the command line and terminal interfaces. Keep branch reviews in the linked worktree where they were started. 2026-08-03 to 2026-08-09
12 PRs merged (10 Kenn, 2 community) Weekly diff: +18,220 / −3,493 lines
Run roborev CI reviews on GitLab merge requests. Use Grok Build as a first-class coding agent. Track CI costs at the individual job level. Snooze agent hooks for a specific workspace. Daemon discovery now works in sandboxed environments, and interrupted work resumes after agent Stop hooks. 2026-07-27 to 2026-08-02
6 PRs merged (2 Kenn, 4 community) Weekly diff: +3,958 / −853 lines
Configure multiple named ACP agents and use Goose for reviews. Keep completed-job prompts and outputs out of job listings. Use REVIEW.md guidelines when no review guidelines are configured. Treat empty Antigravity reviews as failures instead of completed reviews. See commit ranges rendered correctly in the CLI instead of as truncated commit IDs. 2026-07-20 to 2026-07-26
8 PRs merged (7 Kenn, 1 community) Weekly diff: +705 / −77 lines
Post-commit hooks no longer enqueue duplicate reviews for the same commit. PostgreSQL sync URLs can reference a password stored in a file. Agent checks now honor configured commands. Reviews avoid recommending tests that merely repeat the implementation logic. 2026-07-13 to 2026-07-19
17 PRs merged (15 Kenn, 2 community) Weekly diff: +4,111 / −193 lines
Automations can now capture machine-readable receipts when review runs launch. CI quiet hours now throttle runs for all users, with selective bypasses for work that should proceed. Reviews queued from a detached HEAD now infer the branch, and the terminal UI shows a clear placeholder instead of leaving it blank. Custom installation paths are now supported for agent skills. Queued jobs now show their prompt correctly with roborev show --prompt. 2026-07-06 to 2026-07-12
19 PRs merged (12 Kenn, 7 community) Weekly diff: +6,584 / −165 lines
Automatic commit reviews work reliably again because managed Git hooks are repaired when the daemon starts. Cancel an individual review job with the new roborev cancel command. CI panel terminal metrics now persist and can be exported with ci-metrics. Claude Code and Codex skills now run only when explicitly requested, and skill installation respects CLAUDE_CONFIG_DIR and CODEX_HOME. Gemini and other ACP agents now receive the correct paired model settings, including fixes for Antigravity print-mode reviews. 2026-06-29 to 2026-07-05
23 PRs merged (23 Kenn, 0 community) Weekly diff: +10,326 / −1,760 lines
Export completed reviews incrementally using cursors. Use roborev with Factory Droid through new hook and skill support. Automatic commit reviews recover when managed hooks break after an update. Post-commit hooks are more reliable and configurable for large repositories on Windows. Job listings load metadata without including full prompts, and review panels show wall-clock elapsed time. 2026-06-22 to 2026-06-28
32 PRs merged (31 Kenn, 1 community) Weekly diff: +7,734 / −466 lines
Windows commands no longer hang on terminal color probes or open stray console windows, and summary filters and Codex token usage now work correctly. Copilot JSON review output is parsed correctly, preventing truncated output from causing false failures. Each analysis type can now use its own agent configuration, and a new lookahead review detects time-series bias. Codex users can pass configuration overrides with `-c`, and fix commits support configurable metadata. Updates can complete over slower connections instead of timing out after 30 seconds. 2026-06-15 to 2026-06-21
19 PRs merged (17 Kenn, 2 community) Weekly diff: +5,666 / −431 lines
Update checks now keep working when GitHub API rate limits are reached. Windows users no longer see flashing console windows from daemon child processes. Review guidelines can now be configured globally. Agent health checks now use resolved command names, including aliases such as `agy` for Gemini. Agent quota cooldowns now respect configured limits, while CI retries stay on the configured agent. 2026-06-08 to 2026-06-14
20 PRs merged (19 Kenn, 1 community) Weekly diff: +16,563 / −859 lines
Pause and resume the review queue without stopping the daemon. The TUI queue now starts and renders faster, especially for larger queues. Windows daemon restarts no longer fail or open unwanted console windows, and updates work in PowerShell 5.x. Review cost tracking now shows aggregate costs. Agent hooks can use a custom binary and filter by branch. 2026-06-01 to 2026-06-07
27 PRs merged (26 Kenn, 1 community) Weekly diff: +33,243 / −12,608 lines
Run multiple review agents in parallel and compare their findings in a review panel. Use optional agent hooks for automatic reviews, including from detached Git worktrees. CI reviews now read the current checkout and retry safely instead of reviewing stale code. The TUI no longer freezes when filtering across multiple repositories. Compact reviews now require actionable findings and produce fewer false positives from dependency metadata. 2026-05-25 to 2026-05-31
10 PRs merged (10 Kenn, 0 community) Weekly diff: +3,362 / −1,347 lines
Classifier availability checks now work correctly, with added support for Pi schemas. Hooks now find the correct roborev binary in managed installations. Update checks now avoid GitHub API rate limits, and reviewer version checks use the project's toolchain. The TUI can toggle full command lines in log and prompt views. OpenCode installation now uses the supported source, Copilot streaming is disabled when supported, and CI batches have more time to finish. 2026-05-18 to 2026-05-24
13 PRs merged (12 Kenn, 1 community) Weekly diff: +3,154 / −1,188 lines
The queue view now shows estimated agent costs. Gemini reviews now prefer the Antigravity CLI. Retried jobs are deferred until their next eligible retry time. CLI invocation errors now show usage help, while runtime failures show a concise error. Git operations now support worktrees backed by bare clones. 2026-05-11 to 2026-05-17
9 PRs merged (9 Kenn, 0 community) Weekly diff: +1,693 / −154 lines
Run `roborev analyze` in the new built-in security mode to inspect existing files and codebases. Find the daemon endpoint programmatically in `roborev status --json` output. See a cleaner TUI queue, with automatic design-router activity hidden by default. Gemini agents can now access diff snapshot directories during reviews. Codex review sessions resume correctly with the updated command arguments. 2026-05-04 to 2026-05-10
15 PRs merged (14 Kenn, 1 community) Weekly diff: +4,058 / −759 lines
Run roborev fix in bounded batches with --batch-size, then continue interrupted work with --resume. Foreground fix runs now stop when the agent hits a quota or session limit. Branch base detection now handles URL-based upstreams and stale local base branches correctly. Review findings are displayed separately instead of appearing as one continuous block. Large diffs remain outside prompts, reducing problems when reviewing substantial changes. 2026-04-27 to 2026-05-03
7 PRs merged (7 Kenn, 0 community) Weekly diff: +8,872 / −2,356 lines
Renamed repositories now appear correctly when filtering in the TUI. roborev fix can now discover reviews while the repository is in detached HEAD state. Claude Code scheduled task files are now ignored during reviews. Review type flag completion now includes all available review types. 2026-04-20 to 2026-04-26
3 PRs merged (3 Kenn, 0 community) Weekly diff: +8,072 / −146 lines
You can opt in to automatically route design reviews. Branch reviews now use the branch’s configured upstream instead of the default branch. 2026-04-13 to 2026-04-19
11 PRs merged (8 Kenn, 3 community) Weekly diff: +4,415 / −1,199 lines
Claude Code can now connect through a proxy using the <model>@<base_url> format. Copying from the terminal interface now works over SSH using OSC52. Filter searches now accept j, k, and q without triggering navigation or closing the modal. Reviews are more consistent and produce less noise through improved prompts. 2026-04-06 to 2026-04-12
9 PRs merged (9 Kenn, 0 community) Weekly diff: +4,390 / −1,704 lines
API integrations can now use roborev's OpenAPI support. Codex reviews now work in read-only sandboxes, including oversized diffs passed through a unified diff file. Review results can now be filtered by review level, with minimum severity cascading through the selection. roborev fix no longer picks up jobs from unrelated branches. OpenCode output now renders tool calls and omits migration noise from stderr. 2026-03-30 to 2026-04-05
23 PRs merged (16 Kenn, 7 community) Weekly diff: +7,574 / −1,344 lines
Reviews now consider pull request discussions and per-commit feedback, while automated fixes include user comments and previous tool attempts. The terminal interface updates instantly, preserves cursor position, rejects stale responses, and sizes columns correctly. Global reasoning defaults now apply consistently, with fixes for configuration resolution across queued jobs, reruns, and worktrees. Passing reviews can now close automatically with the new auto_close_passing_reviews option. Linux users can use XDG_RUNTIME_DIR for daemon sockets, and downstream packagers now have systemd unit files. 2026-03-23 to 2026-03-29
25 PRs merged (21 Kenn, 4 community) Weekly diff: +10,745 / −5,751 lines
Codex reviews handle large diffs more reliably with improved prompt budgeting and fallbacks. roborev now supports systemd socket activation. Copilot reviews no longer lose permissions in non-interactive mode. TUI selections stay in place while jobs and reviews refresh, including in distraction-free mode. Branch refinement now uses the requested review agent, and reruns preserve the requested model information. 2026-03-16 to 2026-03-22
27 PRs merged (26 Kenn, 1 community) Weekly diff: +11,037 / −2,996 lines
See aggregate review statistics with the new roborev summary command. Control the TUI through a local socket, keep it running with --no-quit, and inspect runtime metadata. Run daemon communication over Unix domain sockets. Track token usage through agentsview, with version checks to avoid unsupported calls. Review agents now use a read-only sandbox, preventing .git/index.lock conflicts; worktree branch filtering also works correctly. 2026-03-09 to 2026-03-15
18 PRs merged (18 Kenn, 0 community) Weekly diff: +18,057 / −20,002 lines
Post-commit reviews now run correctly in linked Git worktrees, and roborev fix no longer picks up unrelated reviews there. Reviews now skip common lockfiles by default, with configurable repository-level exclude patterns. Custom agent command settings are now honored when checking whether an agent is available. The terminal UI now supports clickable column options, wraps long review comments, and can show session IDs in the queue. Codex users can select the maximum reasoning level, and skill names now work with GitHub Copilot. 2026-03-02 to 2026-03-08
50 PRs merged (46 Kenn, 4 community) Weekly diff: +18,999 / −3,676 lines
Review multiple repositories from one workspace, with optional automatic filtering to the current worktree. Run reviews with Kiro, Cursor, and pi coding agents, with more reliable agent and model selection. Configure CI review matrices, throttle redundant runs, and cancel reviews for closed or superseded pull requests. Target fix and refine commands by minimum severity, skip commits by message pattern, and save patches to disk. The terminal interface now handles navigation, quitting, embedded terminals, mouse settings, and close failures more reliably. 2026-02-23 to 2026-03-01
43 PRs merged (39 Kenn, 4 community) Weekly diff: +59,820 / −42,980 lines
Use more coding agents through Agent Client Protocol support and the new Kilo CLI integration. Inspect persistent job output in a unified TUI log viewer, with a new daemon activity log for troubleshooting. Run fixes directly from the review view, with reliable worktree handling, preserved review context, visible OpenCode output, and PASS reviews skipped automatically. Use the TUI more comfortably with mouse interactions, copy and paste, improved spacing, compact layouts for short terminals, and a distraction-free mode. Wait for multiple jobs at once, target a repository or branch when opening the TUI, and use expanded shell completions for agents, reasoning levels, and analysis types. 2026-02-16 to 2026-02-22
26 PRs merged (23 Kenn, 3 community) Weekly diff: +35,257 / −17,112 lines
Run reviews in CI with the new `roborev ci review` command and generate a GitHub Actions workflow with `init gh-action`. Trigger fixes from the TUI in background worktrees, now created from the reviewed commit instead of the current HEAD. Keep review history across clean rebases, and use `roborev compact` to verify and consolidate findings. Review jobs can fall back to a backup agent and pause gracefully when an agent reaches its quota. Large reviews now work around Windows' 32 KB command-line limit, while hook initialization and migrations with quoted table names no longer crash. 2026-02-09 to 2026-02-15
28 PRs merged (22 Kenn, 6 community) Weekly diff: +20,820 / −11,029 lines
Find reviews faster in the TUI with unified filtering, search, lazy branch loading, and current-directory prioritization. Manage settings from the command line with roborev config get, set, and list. Wait for a review to finish in scripts and automated workflows with the new roborev wait command. Analyze named branches and request design or security reviews directly from the command line. Reviews now work more reliably in Git worktrees, refine loops survive git hook failures, and Codex fixes show visible output. 2026-02-02 to 2026-02-08
20 PRs merged (17 Kenn, 3 community) Weekly diff: +7,941 / −723 lines
Browse review jobs with the new `roborev list` command and get structured output from `roborev show --json`. Run automated GitHub pull request reviews with the new CI poller. Use the new `/roborev:fix` skill to address review findings, and check its availability with `roborev skills`. Post-commit reviews no longer run twice or use the wrong agent, and existing hooks upgrade more reliably. The TUI now labels queue entries as JobID and color-codes whether findings have been addressed. 2026-01-26 to 2026-02-01
43 PRs merged (43 Kenn, 0 community) Weekly diff: +20,667 / −10,495 lines
Run built-in code analysis and apply fixes directly, including batch fixes for unaddressed findings. Watch agent output as it streams with the new tail command. Find reviews by branch in the TUI, with branch filtering that works across the full result set. Run reviews without the daemon, use multiple clones of the same repository, and avoid duplicate review jobs. Use Cursor as a review agent, verify configured agents with check-agents, and automate follow-up work with review hooks. 2026-01-19 to 2026-01-25
46 PRs merged (45 Kenn, 1 community) Weekly diff: +21,949 / −1,759 lines
Share reviews across multiple machines with PostgreSQL sync. Choose models for every agent and run reviews with the new Factory Droid integration. Install roborev on Windows with PowerShell, including ARM64 support, or through the new Homebrew tap. Work faster in the TUI with a commit viewer, in-app responses, clipboard copying, clearer feedback, and improved keyboard navigation. Apply configuration changes without restarting the daemon, and limit reviews to recent commits with the new --since option. 2026-01-12 to 2026-01-18
24 PRs merged (23 Kenn, 1 community) Weekly diff: +16,978 / −1,659 lines
Automatically address review feedback with the new refine command, including changes on main selected with --since. Review uncommitted changes with roborev review --dirty, and use --wait to stay until the review finishes. Install ready-made roborev skills for Codex and Claude Code with roborev skills install. Exclude selected branches from reviews and give repositories clearer display names in configuration. Reviews now include full commit message bodies, avoid several false failure results, and keep queue navigation stable when hiding addressed reviews. 2026-01-05 to 2026-01-11
29 PRs merged (21 Kenn, 8 community) Weekly diff: +21,886 / −2,783 lines
The terminal UI now supports repository filtering, pagination, responsive sizing, and hiding addressed reviews. Review details are easier to scan with branch names, PASS/FAIL verdicts, retry counts, and clearer addressed status. You can navigate between reviews, cancel running jobs, and rerun failed, canceled, or completed reviews directly from the terminal UI. roborev now supports Gemini CLI, GitHub Copilot CLI, and OpenCode as reviewers. New update commands, configurable timeouts and data storage, JSONL completion events, and correct worktree handling make roborev easier to operate and integrate.