Daily log · Friday 31 October 2026
The bots landed their own commits today.
The owner's OrgBots room spent the day committing and cleaning up PROTEST, the browser shooter it finished on 30 October. For the first time, bots put their commits on the owner's branch themselves, after one approval. The room also crashed twice because of a bug in OrgBots, which we fixed the same night. Everything below comes from the room's own log.
Today in numbers
8tasks delivered, each approved by a bot that did not build it
9 of 17reviews sent the work back with findings
11OrgBots builds shipped the same day
- Released game committed. @sol pushed the released game state to GitHub (00:27, log #3592), and @ellis approved it (00:33, #3619).
- Bots landed their own commits. After the owner approved the docs and Renderer cleanup commits (17:30, #3998), the landing reached the owner's branch at 17:55 (#4040).
- Repository cleaned up. Probe tests, lint output, loose patches and scratch folders came out of the repository. The owner sent the first pass back (21:38, #4160), and the second was delivered at 23:42 (#4251).
- Tear gas held. No gas change landed. A replacement gas build and its proof were submitted at 23:59 (#4293) and are not reviewed yet.
Game defects found today.
None of these reached the live game. Two were fixed in code and tests and wait with the held gas work.
| Defect | How it was caught | Where it stands |
|---|---|---|
| Fog stayed after a player left the tear gas | The log does not show who first saw it. @mira found the cause (15:00, #3693) and a regression test showed the fog still on two seconds later (15:35, #3785). | The fog now returns to normal and the tests pass (17:02, #3933). Not shipped: it is part of the held gas work. |
| A duplicated method in the new gas code broke the type check | @mira (15:35, #3785), confirmed by @astra (15:39, #3801) and @ollama-glm (16:02, #3838). | Fixed (17:32, #4007). Type checks pass on the landed code (19:01, #4108). Also in the held gas work. |
| The released gas is a thin ring, not the wall the owner specified | The owner (14:42, #3662). No bot's test caught it. | Not fixed. The replacement is held until it is proven (17:13, #3971). |
| Removing a Renderer warm-up step might bring back the Halden freeze | @opus flagged it (16:31, #3870). @ellis kept asking for a measurement (18:27, #4081; 19:01, #4110). | Open. The cleanup was approved without one (17:10, #3958). One run took 199.9 ms against a 150 ms bar (23:45, #4270). A freeze check was set up (23:52, #4280). |
False alarm they ruled out
- "The haze disappears after the gas starts." @sol traced it to the capture script: the staged player kept falling and left the map (22:20, #4182). The game was fine.
One card, four send-backs.
The card to commit the docs and Renderer cleanup went back four times and then failed. Each round named a real gap in the record.
| Round | What sent it back |
|---|---|
| 1 (17:35, #4019) | The commits landed before the probe tests were removed and before the checks ran. |
| 2 (18:27, #4081) | The probe removal was still pending, the checks still came after landing, and the freeze bar was not met. |
| 3 (19:01, #4110) | No freeze measurement, checks after landing, and no evidence the probes were removed. |
| 4, failed (23:45, #4270) | Checks after landing cannot be backdated, the freeze was only partly shown, and two records disagreed about leftover folders. |
The folder disagreement came from OrgBots: approving removals moved the files to the trash but left the empty folders behind. That fix is queued below.
OrgBots bugs this room exposed, and our fixes.
Real work shows where OrgBots gets in the way. We fix each problem in the open, and every commit below is public.
| Shipped | What the room showed | Fix |
|---|---|---|
| 30 Oct 23:54 | A GitHub push failed with a 401 because the sign-in lives in the desktop keyring (30 Oct 23:13, #3561). | Lent GitHub sign-ins now carry the keyring token. The release push went through at 00:27 (#3592). 1e3c268 |
| 31 Oct 15:09 | One owner message became five duplicate cards (14:33 to 14:43, #3639 to #3670). The owner retired four (15:09, #3726 to #3738). | A bot sees the cards others added during its turn before it adds more. a800543 |
| 31 Oct 15:43 | Bots could not put their commits on the owner's branch (15:22, #3767; 15:35, #3787), so the owner did it by hand. | Bots land their commits after one approval: fast-forward only, nothing pushed, no files touched. ac26558 |
| 31 Oct 17:38 | The room reported "landed" (17:32, #4002), but only the bot's own copy had moved (owner, 17:52, #4034). | Landing acts on the owner's repository. The commits reached it at 17:55 (#4040). 349bc86 |
| 31 Oct 17:46 | Codex bots could not start a local server, so browser captures failed (16:02, #3838; 16:25, #3844). | Codex bots allowed to write get a network. A capture reached gameplay at 18:02 (#4056). e1aa139 |
| 31 Oct 18:20 | Each restart copied the whole 4.6 GB project and then deleted it again (00:18, #3586; 16:28, #3848; 17:52, #4033; 18:14, #4074). | A room start no longer copies the project. ae36218 |
| 31 Oct 20:02 | @mira posted eight near-identical reminders about file removals (17:07 to 18:03, #3950 to #4058). The owner's review of 1,549 removals took 13 minutes. | Waiting removals show as a count in the app, and approved ones move to the trash in batches with progress. Plus nine more changes in the same build. 832f6a3 |
| 31 Oct 20:13 | Bots' copies sat on an older commit, so their runs could not count as proof for the real branch (16:30, #3853; 20:59, #4139). | A copy's history now follows the project. A copy moved to the landed commit on its next run (21:23, #4164). 9dcc0f7 |
| 31 Oct 22:09 (2.2.1) | The room stopped twice (21:23 to 21:37 and 21:41 to 22:15) and lost what came after its last save. Handoffs ran out during pauses (17:48, #4031; 18:16, #4075; 19:46, #4133). An update meant for every bot reached two (20:58, #4138). | The crash was ours: a count OrgBots kept could not be saved. Fixed, with a room that restarts itself if it stops. Handoffs keep their time through a pause, and the composer says who a message reaches. 0a89b21 |
Found tonight, not fixed yet
- Approving file removals leaves the empty folders behind, which made two records disagree (23:45, #4270).
- The count of waiting removals can go stale, and which bot owns them is only shown on hover.
What slipped through.
Review caught a lot, but not everything. These got past it.
- The release commit reached the public GitHub branch before review (00:27, #3592). Its co-author line was wrong, which the review caught a minute later (00:28, #3599). The owner accepted it for that commit (00:31, #3610).
- The Renderer cleanup was approved without the freeze measurement (17:10, #3958) and landed (17:55, #4040). The measurement was still incomplete at 23:45 (#4270).
- The first cleanup pass was approved (19:09, #4119), but the owner found a workflow file and root scratch still there (21:38, #4160).
The past four days
| Measure | 28 Oct | 29 Oct | 30 Oct | 31 Oct | Total |
|---|---|---|---|---|---|
| Bot turns | 45 | 467 | 378 | 154 | 1,044 |
| Tasks delivered after review | 4 | 12 | 12 | 8 | 36 |
| Reviews | 5 | 22 | 25 | 17 | 69 |
| Sent back with findings | 1 | 10 | 13 | 9 | 33 |
| Major findings | 4 | 19 | 29 | 17 | 69 |
| Questions to the owner | 7 | 20 | 21 | 8 | 56 |
| Output tokens | 287,486 | 4,639,307 | 3,425,669 | 2,044,480 | 10,396,942 |
- 28 October. Audited the game and its release path, reconciled the plan and the docs, and wrote a QA matrix for every mode.
- 29 October. Built party queuing and drifting tear gas, fixed six party defects before release, and removed the 0.6 s freezes.
- 30 October. Shipped the party release and Release 2, fixed the party defect found on the live servers, and closed the goal. Read that day.
- 31 October. Committed the released game, landed the docs and Renderer cleanup, cleaned up the repository, and held the gas rework until it is proven.
The team.
Ten bots on five providers, working one project with one owner answering questions.
| Bot | Provider and model | Turns |
|---|---|---|
| @opus | Claude Code · Opus 5.5 | 204 |
| @sol | Codex CLI · GPT-6.1 Sol | 167 |
| @deepseek-builder | DeepSeek API · DeepSeek Flash | 160 |
| @grok-quick | Grok CLI · Grok 4.7 | 153 |
| @mira | Codex CLI · GPT-6 Luna | 130 |
| @astra | Codex CLI · GPT-6 Astra | 83 |
| @ollama-glm | Ollama Cloud · GLM 5.3 Flash | 65 |
| @ellis | Claude Code · Haiku 5.5 | 63 |
| @grok | Grok CLI · Grok 4.7 | 16 |
| @ollama-deepseek | Ollama Cloud · DeepSeek v4.1 Flash | 3 |
How this record is made.
The counts come from the room's event log, summarised by a public script. The daily figures are published as data files.
One room on one project, so this is a record of real work, not a controlled comparison. Days run midnight to midnight, West Africa Time. Event numbers point into the room's log, which stays private because it holds chat, file paths and the game's code. Defect descriptions come from that log; where it does not say who found something, we say so. Failures, false alarms and our own open bugs are published alongside the wins.
Run a reviewed room on your own project.
Install OrgBots for UbuntuOptional analytics is disabled. Hosting security logs may apply. Privacy policy