Agent Durbar v0.1.17

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.17 (release, 2026-11-12)

• A helper's button that has the keyboard focus is not retired (spec 005 FR-242, issue #1483): a finished helper whose button held the focus (its head in the Real office with +K open, its tile in the Open office, its row in the list) could be retired by the next end and the focus fell to the page. It is now kept while its own button has the focus (as the block in the chat already was) and retired by the next end once the focus leaves. Checked by check-helper-retention (tile in the flat office, row in every world, head with CHR_WORLD=real; mutants without each part of the rule fail).
• Fixed: UI checks that wait on the window focus no longer fail when another check's window is open (spec 008 FR-210, issue #1498): with the suite's default of two drivers at once, shots-097-screens failed 5 of 5 runs on a Mac (timed out waiting for the window active again) and passed 5 of 5 one at a time. scripts/ui-suite.js now runs entries marked focus: true (shots-097-screens, both worlds, and check-097-hover-card) alone, after the others, when the windows share a desktop (macOS, Windows, Linux with a display); Linux under xvfb is unchanged. Checked by test:uisuite; Linux and Windows runs not done.
• Windows: a .cmd program in a folder with spaces now starts (spec 009 FR-029, issue #1525): VS Code's code.cmd, npm's npm.cmd and other .cmd programs went through cmd.exe /d /s /c without the outer pair of quotes that /s removes, so a path such as C:\Program Files\nodejs\npm.cmd broke at the first space. One shared helper now builds that line for all three routes and still refuses characters cmd.exe could interpret; an update check that is refused this way now says This command cannot be run safely on this computer. instead of exit EUNSAFE. Not run on Windows by the author; Windows CI runs the real case.
• A Markdown-escaped token is hidden on the permission card (spec 012 FR-180, issue #1490): a token written with an escape (ghp\_…) was shown in a permission request's summary, while the desk already hid it. The label mask now hides such a token in a command, a tool request, the activity log and the live readout, and shows the rest as written; when it cannot be sure, the card shows its neutral words.
• An escaped token is hidden everywhere the shared masks run (spec 012 FR-181, issue #1541): a token written with a backslash inside (ghp\_…, g\hp_…, sk\-proj\-…) was shown, whole or in part, in the "Approve for this session" label, the activity log and its export, the brain view's step targets, the MCP config view and copy preview, and the read tools' answers to the agent. These now hide it and keep the rest of the text as written. Text glued after a hidden environment value ([redacted:NAME]_<rest>) is now hidden like text glued after any other mask, also when that text is written with a backslash. The approval card hides a token cut by an escaped line break (\n) the same way. The reply line on a desk and the live readout hide the rest glued after an escaped token (ghp\_…@…) as the other surfaces do. A tool name longer than 256 characters is shown as "tool". Known, not fixed here: the desk's mask slows down on text with very many hidden-value markers (issue #1593).
• The macOS title-bar check expects the Settings title where Settings puts it (spec 052 FR-097, issue #1458): the check wanted the title on the centred column, but since the Settings section list the title lines up with the list-and-page block (x=194 at 1440 wide), on purpose. The rule and the check now say so, and the check also fails a title under the traffic lights. Test-only, no change to the app.
• check-crew waits for the crew scenes before it reads them (spec 077, issue #1460): the dark fixture entry of the Cheeky Crew driver failed on the macOS and Windows CI runners at "grown desks draw portraits (0)" and "crew scenes on screen (0)", and passes on a Mac and on Linux. The driver now waits at most 5 s after picking the crew for the crew scenes to be drawn, then runs the same checks unchanged. The cause on the runners is not proven and the fix is not yet seen on them; it needs a nightly. Test-only, no change to the app.
• Same-file checks compare 64-bit file IDs exactly (spec 083, issue #1491): the notes store, the world store, the machine policy.json read and the agent-file read on Windows now read inode and device numbers as bigints, so two different files whose NTFS file IDs differ only in the last bit (above 2^53) are no longer taken for one. Nothing stored changes. Not run on Windows or macOS; checked with stubbed IDs on Linux.
• In the Retro arcade, the first draw no longer puts a request pill over a name plate (spec 097 FR-358, issue #1434): for about half a second after the office appeared at 1280x800 the limit and inbox pills lay over plates. Checked with check-097-slip-plates on the simulator; not run on a Mac.
• check-097-crews waits for the lobby's visits to settle before it reads the log (spec 097, issue #1459): the Characters driver failed on CI at sets/rebels/owl-runA.webp 0 net::ERR_FAILED (Linux) and at an unloaded picture (macOS), while its own load skipped the wait for the lobby's visits. It now waits at most 5 s for them, checks the scene did not change before it reads, and prints request times when a file fails; ERR_FAILED is still a failure (a mutant that injects it into the log must fail; a real one is read: only). The cause on the runners is not proven and the fix is not yet seen on them. Test-only, no change to the app.
• Pointer log for two more hover-card steps (spec 097, FR-328): the simulator driver check-097-hover-card now records the pointer events of its asleep-card step and of its third-agent held-card step and prints them when one fails, like its first pointer step. Diagnostics only: no wait, limit or verdict changed. Checked by test:pointer-log (the wrapping in the driver's source) and a forced failure.
• Desk-feature checks also run in the Retro arcade (spec 102, issue #1399): test drivers only; five desk-feature checks (hover card, desk marks, one face, chat look, chat contrast) now also run in the Retro arcade in the UI suite. Nothing changes for users.
• Spec note only: the arcade layout check with the chat open (check-office-overlap with OFFICE_WORLD=retro, already in the arcade job of check-parallel) is recorded as done in spec 102 (T261); T225 is pointed to PR #1528 and the acceptance wording of #1399 is written as an owner question. No code change.
• Pictures of the Retro arcade at 12 and 15 agents, in rooms and with one room each, at both window sizes (spec 102 T281, issue #1428): looked at on macOS in the simulator; the layout check's picture option no longer leaves a tooltip over the cabinets. At 12 and 15 agents at 1280x800 the plate keeps its helper count and dots, as the spec says.
• shots-102-arcade reads a person only once it has a box (spec 102, issue #1517): the arcade driver failed once on Linux CI (run 42966366604) with IndexSizeError ... getImageData ... source width is 0 in the "every person whole beside the cabinet" check, a person whose box rounded to 0 px. The check now reads such a person again for the same bounded 5 s an unloaded picture gets, and if it stays so fails naming the person, its size and its style. The cause of that one red is not established and the fix is not yet seen on Linux CI. Test-only, no change to the app.

Agent Durbar v0.1.16

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.16 (release, 2026-11-11)

Highlights

• Long chats stop growing without limit in memory: a chat keeps its newest 300 messages and an agent its newest 50 finished and 20 failed helpers in full (older ones shrink to one line), with a note at the top of the chat saying how many are not kept in this view.
• A Claude session's instructions stay out of the process list: the folder's CLAUDE.md text, agent profile and hooks go through private files instead of the command line.
• Screen-reader announcements mask tokens and home folders: announcements now use the same masking as the desk (toasts and some labels are still not covered, see Known issues).
• "Delete the file" for the project notes is much harder to trick: it deletes only the file Durbar wrote, though a rare swap by a local process can still get through (see Known issues).
• Auto-follow no longer brings back an old "finished" after you send the agent another message.

Known issues

• Project notes file (issue #1464): a local process that swaps a folder link at the right instant can still make Durbar move another folder's notes.md into your project under a random name (the message names it), a delete with no inode record (file written by an older version, or on a drive without inode numbers) can still remove it, and a write can still overwrite it. This is recorded in docs/safety.md.
• Announcements (issue #1320): toasts, the chat's label and the notice strips still show an agent or folder name as written.

All changes

Memory and performance

• A long chat no longer grows without limit in memory (spec 005 FR-238, FR-239, issue #1380): every chat kept every message and tool call while the app stayed open, so the window grew by about 35 MB an hour in the simulator. A chat now keeps its newest 300 messages, the newest 300 steps of a turn or helper still running, and at most 2,000 steps in all. Your last question and the last reply are always kept, as are waiting questions and permission cards and the live parts of a turn. One line at the top says how many earlier messages and steps are not kept in this view (the agent app's session keeps its own record). While you are scrolled up in an open chat nothing is removed, and what you focus or select, or a note saved from a message, stays. Checked with test:chattrim and the check-chat-memory simulator driver (Open office, Real office and Retro arcade); Linux only.
• Finished helpers no longer grow without limit in memory (spec 005 FR-240, FR-241, issue #1418): an agent now keeps its newest 50 finished and 20 failed helpers in full. Older ones leave the office and the helper lists, and their block in the chat becomes one line with the helper's name, result and status. A helper that is still running, the one whose activity you have open, and one you are focused on or selecting text in are never removed. Checked with test:subretire and the check-helper-retention simulator driver (three worlds, 1440x900 and 1280x800); Linux only.

Agent follow and chat

• Auto-follow no longer brings back an old "finished" (issue #1444, the known issue of v0.1.15): if an agent finished while its chat was on screen and you then sent it another message, the window could go to the office and then reopen the chat with <name> finished, although the agent was already working on your message. A reply you saw in its chat now counts as seen, and a new message replaces a completion still waiting. Checked in the simulator on Linux.
• A chat keeps its place across the Browser tab (spec 005 FR-235, issue #1168): a chat that was following the end is at the end again, and still following, when you come back from Browser to Chat, also when the agent wrote meanwhile. A chat you had scrolled up keeps its position. The chat now keeps its width while collapsed, also where scrollbars take room (Windows, Linux). Checked on macOS only, with a simulated classic scrollbar, at 1440x900 and 1280x800; not on Windows or Linux.
• On a Mac, the top of the rail's resize handle resizes the rail (spec 005 FR-236, issue #1457): its top 40 pixels used to move the window instead. Checked in the simulator; not on a real Mac.

Safety and privacy

• A Claude session's instructions and settings no longer show in the process list (spec 012 FR-179, issue #1379): the folder's CLAUDE.md text, a chosen agent profile and the folder's accepted hooks and rules used to go to Claude Code on its command line, which other accounts on the same computer can read. They now go through private files that only your account can read and that are removed when the session's Claude process ends. A Claude Code too old to read them (before 2.1.69 for CLAUDE.md, before 2.1.281 for agent profiles) gets a message asking you to update it. Checked with unit tests on a fake Claude process (macOS); not with a real Claude session or on Windows.
• Screen-reader announcements no longer read out a token or a home folder (spec 012 FR-178, issue #1419): lines such as <name> finished. New reply. used the name as written. They now go through the same mask the desk already uses. A command or question in a request is said exactly as the card shows it, and "needs your OK" is said even when the request is hidden. When masking is not possible, a fixed line says a line was not read out. Toasts, the chat's label and the notice strips still show the name as written (#1320). Checked with test:announcemask and one simulator run; no screen reader run.
• Deleting the project notes file deletes only the file Durbar wrote (spec 083 FR-055, issue #1463): a folder link swapped within microseconds by a local process (for example a sandboxed agent) could make Delete the file remove a notes.md in another folder, or make the way back replace one. Durbar now records the inode of the file it writes and deletes only that one (a notes.md with the same text but another inode, after a git checkout, is left as changed). The way back is a hard link, or on drives without hard links a copy into a new notes.md that never replaces a file. Part of the problem is still open (see Known issues, #1464). Checked by unit tests with injected file calls and a real swapper process; Linux only.

Copilot on a Mac

• Playwright MCP can start again under Copilot on macOS (spec 087 FR-024, issue #1173): Copilot's private temp folder is now a short one under /tmp when the system temp folder is too long for a Unix socket (macOS /var/folders). The folder is still private to your account and removed afterwards; Windows is unchanged. Checked on Linux; not on a real Mac.

Behind the scenes (nothing changes in the app): the team merge gate reads CodeQL results and a PR's changed files more strictly (FR-090, FR-091, #1422, #1479), the UI check runner now ends a timed-out check's display and window whole (FR-199, #1417), the live-activity UI check re-reads the lobby until it settles (FR-191, #1481), and the README build-cost sentence says how many machines started on each day (#1387). Checked by script tests only; not against live GitHub, not on macOS or Windows.

What was checked

Linux and the simulator for most changes, as noted per line. macOS: the tab-follow check only, plus unit tests for the private launch files. Windows: not run by hand for any change here. No real agent session or screen reader was used for these notes.

Agent Durbar v0.1.15

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.15 (release, 2026-11-10)

Highlights

• Auto-follow tells you before it moves you: a notice says where it is taking you and why, and counts down 3 seconds; Esc cancels.
• Lookalike letters flagged on approval cards: a name that mixes alphabets (a Cyrillic letter in a Latin word) gets a visible marker and a plain warning line.
• Real office: every character crew draws its own people, desks and rooms can be rearranged by drag, keyboard and menu, and helper heads stay on one row.

Known issue

• Auto-follow notice (issue #1444): if the agent whose chat you are in finishes while you are typing, then after the notice moves you to the office a "finished" notice can reopen that chat after 3 seconds (it already happened as a quick flicker in v0.1.14); Escape cancels it; fix is in review for v0.1.16.

All changes

Agent follow and chat

• Auto-follow tells you before it moves you (spec 052, PR #1414): with Auto-follow on, a full-screen notice now says where it is taking you and why ("Priya is working on this", "Priya finished. See if it is OK.") and counts down 3 seconds before the window goes to the office or opens a chat. Stay here / Stay in the office or Esc cancels the move. Screen readers hear it once; with reduced motion nothing in it moves. Checked in the simulator in Open office, the Photo office, the cave and the arcade at 1440x900 and 1280x800. On macOS only its check script was run after it landed; that run found the script needed a Mac fix (PR #1443, merged).
• Accessibility (spec 005 FR-237, issue #1324): a screen reader no longer hears "Working…" instead of " needs your OK" or a question when both are announced in the same instant: a request outranks a status.

Safety

• Lookalike letters on approval cards (spec 003 FR-113..FR-119, issue #1300): a host, program or MCP tool name that mixes scripts (a Cyrillic letter in a Latin word), uses a styled or fullwidth form or mixes digit systems keeps every letter and gets a visible marker after the odd one, and the card adds one plain line saying so, with an address's host as the network reads it (xn--…). Allow for this session now names exactly what it grants: the whole command, the punycode host, the server and tool. Grant chips and the Why was this allowed? sentence say Has lookalike letters. A warning only: every button stays. Checked in the simulator at 1440x900 and 1280x800; not on a real session.
• CLAUDE.md imports checked by their text first (spec 012 FR-177): the Claude adapter checks each @path import of the folder's CLAUDE.md against the folder by its text before any file is read, checks links inside the folder before they are followed, then checks again once resolved; a path that cannot be resolved is skipped. Checked by tests with Windows and POSIX path cases; not run on a real Windows machine.
• A tool name that looks like a secret is not kept in the usage file (spec 020 FR-033, issue #1291): it is counted under [hidden tool] (a server name under [hidden server]), and an existing file is cleaned when the app starts, with its counts kept. Checked with unit tests, no real session.
• Profile confirmation matches what you saw (spec 043 FR-044): a profile from a folder's repository is trusted only for the exact file text shown in its confirmation; if the file changes before you confirm, the dialog is shown again with a note, and Claude Code sessions start with exactly the text that was confirmed. Linked agent files are not used. Checked by tests; not in the simulator window or on a real session.

Real office

• Every crew in Real office (spec 097 FR-335, FR-336, issue #1154): the Characters choice Cheeky, Pirate, Monster, Royal and Rebels now draws that crew's own people in every Real office world, and Settings says "Real office draws the people.". Minimalist and My characters still show the world's own people.
• Rearrange desks and rooms (spec 097 FR-345..FR-348, issue #1152): drag a desk by its name plate or a room by its folder sign (Esc cancels, a tiny press is still a click); Alt+arrows on a name plate (Alt+Shift+arrows for its room) and the folder menu's Move earlier / Move later work too, on the same saved order as the Open office.
• Helper heads stay on one row (spec 097 FR-338, bug #1384): when the heads do not fit, the last place is +K; no head wraps onto a second line any more (the known issue of v0.1.14).
• A narrow room's sign keeps its + and menu (spec 097 FR-356, issue #1392): in the cave at 1280x800, a one-desk room cut off its sign's + and ⋯ buttons; it now shows the short sign, with the path and cost on the name's tooltip.
• The rename pencil on an arcade cabinet (spec 097 FR-357, bug #1427): it showed at the floor's top-left corner and did nothing; it now sits at the plate's right end and the name field opens over the cabinet's marquee.
• Retro arcade follows the other worlds' rules (spec 102 FR-307..FR-310, issue #1428): big screens with 4 or fewer agents, and its blinking and typing motion starts 2 seconds after the room is drawn. The arcade is not offered yet, so nothing changes for anyone today.
• Minimalist crew art (spec 097, T245): six new drawn people are in the repository, owner-approved; not selectable yet.

Looks and contrast

• Theme contrast checks every colour form (spec 070 FR-080, issue #1259): colours written as rgb(), hsl(), a name or with transparency are now measured and repaired like hex ones; one that cannot be measured is reported as unmeasured. Checked by unit tests and the dark-fixture UI check.
• Readable chips and badges in the Retro arcade (spec 070 FR-081, issue #1278): attachment chips, ask tags, selected rows and tone-coloured badges were 1.2 to 2.7:1 on dark fills; they are now at least 5.9:1. Badges coloured by the status pair are not part of this fix.

Build, CI and team tools (nothing changes in the app)

• A GitHub outage of the Electron download no longer fails every CI job: a workflow keeps the download cache filled for all three systems and builds check the Electron zip offline against the lockfile (spec 008 FR-195).
• Pull requests that change no UI code no longer start the three-system UI checks; main, the nightly and the merge queue still run them (spec 008 FR-191).
• Parity checklists (#1147, #1404) tick themselves when a listed issue closes as completed (spec 021 FR-088).
• The merge gate accepts a CodeQL check that GitHub skips on a PR with no code (spec 021 FR-089, issue #1420).
• Test fixes: the colour parser growth check on Windows (spec 070, T049) and the Real office hover-card check's window focus (spec 097, issue #1408).
• Docs: the feature inventory matches the code (spec 025 FR-021, #1398); spec 080 records a macOS dialog behaviour with numbers for later agent browser control (issue #1415).

What was checked

Linux and the simulator for every change, as noted per line. On macOS the Mac session ran the full UI suite (174 of 174) at commit 8172237 and the arcade pictures and overlap checks at 742144c; the Auto-follow notice (PR #1414) landed after those runs, and on the Mac only its check script was run (it needed the Mac fix in PR #1443, now merged). Windows: CI checks only, no hands-on run. No real agent session was used for these notes.

Agent Durbar v0.1.14

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.14 (release, 2026-11-09)

Highlights

• Real office shows each agent's context on its monitor frame: a thin line fills round the frame and turns red above 90 %.
• Helpers sit at the desk in Real office: each sub-agent is a small head with a status dot under the name plate.
• More secrets masked: short and several-word passwords, mysql -p..., cookies and more stay hidden.

All changes

Real office

• Context on the monitor frame (issue #1148): a thin line round each desk's monitor frame fills clockwise from the top-left as the agent's context fills. It is soft white up to 90 % and red above 90 % (the same steps as the chat header's ring); it draws nothing while the context is unknown, never tints the screen, and does not move. A screen reader hears "context N %" with the desk's name. In the Photo office, the cave and the arcade; 15 agents still fit at 1440x900 and 1280x800.
• Helpers at the desk (issue #1139): each desk's sub-agents show as small round heads with a status dot under the name plate in the Photo office, the cave and the Retro arcade (at 1440x900); click one to open its activity view, +K shows the rest, and the name is in the tooltip. Where there is no room the plate keeps its "+N". 15 agents still fit both window sizes.
  • Known issue (issue #1384): a desk with six or more helpers can wrap the sixth head onto a second line instead of showing +K. Fix planned for the next release.
• The arcade's drag-a-file overlay is readable (issue #1153): it had light text on a light fill; it now uses the page surface, and a new check keeps each world's own colours whatever theme is picked.
• The arcade's name plates show their whole line (issue #1303): a lounge plate's text was cut at 12 and 15 agents; the plates now have room for the full name.

Safety and masking

• More secrets masked (issue #1227): short passwords (4 to 7 characters with a letter and a digit) after a secret name, several-word passwords, Bearer values after a short first word, mysql -pPASSWORD (also across continued lines), AWS secret key names written with spaces, Cookie values without =, and unfinished key starts are hidden; ids, versions and algorithm names (HS256, tls13, v1.2.3) stay readable. Checked by tests only; no real service was called.
• Network home folders hidden in helper names: a home folder on a network share (\\server\share\name, also with extra backslashes, or //server/homes/name) no longer shows its user name in a helper's name, task or tooltip, in Real office and the flat office.

Other

• Coming next, not visible yet: the code for a fourth agent mode, Autopilot, is in place behind a switch that stays off, so the mode menu has no new item and nothing changes for anyone. One stricter check applies today: remote control checks its mode again after its own dialog. Tested on fakes only, not in a real session.

Agent Durbar v0.1.13

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.13 (release, 2026-11-08)

Highlights

• Real office zoom shows model, usage, cost and context.
• The hover card opens over the whole desk, with the agent's last lines, and for a resting agent the session's changes.
• A section list in Settings to jump around the page.
• Masking fixes: tokens, home folders and secret-looking tool names stay hidden in more places.

All changes

Real office

• Model, tokens, cost, context and the last tool in the zoom (issue #1148): hovering a monitor (or tabbing to its desk) shows, under the agent's name, the same facts the Open office desk shows: Model, Now or Last tool, Usage, Cost, Context and time on the task. Unknown facts are left out, and every line is masked like a screen line.
• The hover card opens over the whole desk (issue #1151): hover a person, desk or name plate, not only the monitor. An asleep agent has its card too, with its last reply; a resting agent's card shows the session's changes ("6 files changed, +120 −35, 1 commit, 40 min · counted by Agent Durbar"), as in Open office.
• A desk's unsent draft (issue #1149): a small Draft tag sits on the name plate while a chat has text you have not sent, and the plate's name says "unsent draft", as in Open office.
• The Characters crews, said honestly (issue #1154): Real office draws the Office people for every Characters crew that has no Real office pictures yet, and the line under Office layout says why. Prehistoric now draws its own people in the Photo office.

Settings

• A section list in Settings: a list on the left jumps within the page, marks the section you're in as you scroll, works with the keyboard and in every world, and becomes a row in a narrow window.

Safety and masking

• Masking fixes (issues #1237, #1298, #1295, #1296): two tokens of different kinds written back to back are hidden whole; a secret name split by quotes ('pass'"word"=...) is hidden; the activity log hides a command, tool name or token split by an escape the way the chat does; the chip of an "approve this exact command" scope is masked before it is shortened.
• A permission request never shows a secret-looking tool name or command (issue #1290): the Needs your OK card, its buttons and tooltips, the grant chips and list, and the "why was this allowed" text show a name like a token as tool, and a command keeps its shape with the secret hidden. A card offering "Allow for this session" says when a part is hidden and that the grant covers it. Escape codes are shown as visible markers, and a request Durbar cannot check has no Allow button. What you approve is unchanged.
• A tool named like a secret is hidden (issue #1284) on the chat, desk, helper card and Settings screens; the chip keeps its icon and label.
• A helper's task no longer shows a token or a home folder in the Sub-agent activity panel (issue #1330), its title or the office helper tile (also network \\server\share\user paths).

Other

• Conversation search no longer drops a long message because it holds an emoji.
• The Retro arcade's plan chip is readable (contrast 2.99:1 to 13.14:1) and follows the world's colours.
• README for first-time visitors (issue #1256): the Real office leads the page, with a one-line problem statement, a download line, "Works with / Not yet" and a shorter safety lead. The release clip films the Real office.
• Build cost: the README shows an estimate of what building Durbar with AI agents has cost so far (API list prices, not billing), with a daily chart. Method and privacy: docs/build-cost.md.
• Coming next, not visible yet: the Needs you panel ("Your door") is built behind a switch that is off; five new Real office crews (Pirate, Royal, Rebels, Monster, Cheeky) are in, approved, and wait to be wired up; the first piece of the AI character maker is in the code.

Agent Durbar v0.1.12

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.12 (release, 2026-11-07)

Highlights

• A lounge row in Real office: finished, read agents give their desks to the busy ones.

All changes

• Drag the agent list's width (spec 005 FR-236, issue #1223): the right edge of the rail beside a chat is a splitter. Drag it, or focus it and use the arrow keys (Shift or PageUp/PageDown for bigger steps, Home and End for the smallest and largest); double-click goes back to 220 px. It stays between 180 px and 40% of the window, is remembered on this computer and is never part of a session. Checked in the simulator at 1440x900 and 1280x800.
• The release clip's step captions are numbered 1, 2, 3... in the order they are shown (spec 008; the v0.1.11 clip read "3." then "2."). Checked by test:releasevideo with a mutant; the clip was not re-filmed.
• Agent rules split (spec 021 FR-081, FR-082): AGENTS.md is now a shorter core every agent loads, and task-only rules (incident triage, merge mechanics, review rounds and pace, chat and decisions, keeping docs in sync) moved word for word to docs/rules/, each linked from AGENTS.md; a new priority order (bugs, existing-world parity, the Retro world, new features, other worlds) is in the core. Checked with npm run rules:check (per-file budgets, every task file linked) and a line-by-line verbatim check.
• MCP: the chat's Sign in... button now starts the sign-in (it used to only open the MCP list and showed nothing when the row had no button), also in an idle chat, and the MCP row shows Sign in for a server whose automatic renewal gave up even while Copilot still says it is connected (spec 058 FR-229, issue #1229). Checked with fake-adapter tests; not on a real server yet.
• Notes: Open source (spec 083 FR-046, issue 699): a note's ⋯ has Open source. It opens the chat the note came from (a past session through Previous sessions, sending nothing), scrolls to the message inside the chat and outlines it for a moment; if the session is gone or the message is not found, a toast says so and the note stays.
• Real office: finished agents sit in a lounge row (spec 097 FR-246..FR-249, issue #1220): an agent that is done and whose reply you have read leaves its full desk for a compact row under the desks, seated on their chair beside a small desk and monitor with their name plate; asleep agents too, lamp and screen off. An unread reply keeps the full desk until read. The freed room goes to the working and waiting desks; in the Retro arcade the seats are stools beside a small console. Moving back to work eases (none with reduced motion). Who moves is the flat office's lounge rule. Simulator-checked at 1440x900 and 1280x800.

Build cost

Build cost (API list-price estimate, not billing). Tracking starts today; from tomorrow each release will show daily per-model numbers from every machine. To date, the Linux Claude Code session (running since about 2026-10-25, 318 h): about US$3,758 at API list prices; cost by model Opus 5.5 76%, Sonnet 5.5 24%, Haiku 5.5 and Fable 5.1 under 1%. Tokens: 36.8M input, about 31.9M output, 12.3B cache read, 310M cache write; 97% cache hits. Other machines and teams will be added as they report.

Agent Durbar v0.1.11

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.11 (release, 2026-11-06)

• Open office: the monitor screen shows who needs you, as in Real office (spec 001 FR-230, FR-231, owner decision): a desk's screen turns amber when its agent needs you, orange at the spending limit or on a coffee break, red on an error, and green when a finished agent's reply is not read yet. The glow around the desk card and the waiting halo behind the screen are gone; the status word and the New badge stay. All three desk styles (office, pixel, crew) do it.
• The nightly macOS UI checks press the Mac's own keys (spec 008, issue #839): the answer-key and paste checks use Command on a Mac, as the app expects, and the Real office glass check no longer takes the CI machine's Reduce transparency setting for a bug. One check (characters) now explains itself when it times out. No change to the app.
• Simulator About version from package.json (spec 008 FR-179, #1162): the simulator's About label is the app version plus (simulator), generated at build time, so a release bump edits only package.json and package-lock.json. Checked by test:simversion (with a hard-coded-version mutant).
• A Cookie: or Set-Cookie: header with one opaque token is now hidden (spec 012 FR-168, issue #1159): redact() hid Cookie: name=value pairs but showed Cookie: <long token> whole. A single token of 8 or more characters is hidden, also after pairs, with Set-Cookie attributes kept; prose like cookie: my session bug stays. Checked on Linux by npm run test:runtime-redact; not run on Windows or macOS.
• The merge gate no longer reads "condition holds" as a hold (spec 021 FR-083, issue #1190): scripts/pr-gate.sh matches on hold, owner hold, do not merge, don't merge, hold lifted and off hold only as whole words, so text like "the condition holds" or "threshold lifted" no longer holds or lifts a PR; real phrases still do. Checked by npm run test:trustedcontent (fake gh) on Linux; not run on Windows or macOS.
• Windows UI checks steadier (spec 052, issue #1179): check-auto-follow and check-edit-popups no longer fail on the first try when a neighbour driver's window takes the focus on the shared Windows desktop. Test driver only, no app change; checked on repeated Windows CI runs.
• Copilot MCP sign-ins renew themselves at once after an expiry (spec 058, issue #1201): when a call to a remote MCP server fails with a sign-in error, or Copilot reports the server needs a sign-in after it was connected, Agent Durbar reconnects that server immediately in the background (it waits only for another call to the same server in that chat), with Renewed 12:40 on the server's row. If it worked, the agent gets one short automatic message to retry the step that failed, naming the server; that is one extra message and uses AI credits, and it is never sent at a spending limit, after Stop, or with Refresh MCP sign-ins automatically off. Your own message is never resent, and there is no timer. Probed on a local test server only; not checked on a real Entra server, a real model, macOS or Windows.
• Copilot MCP sign-ins are checked in the background before every message (spec 058, issue #1205): beside each message you send, and when a chat starts, Agent Durbar asks every remote sign-in server that reads connected for its tool list (a real round-trip, no AI, no credits; your message is never held; a server with a successful call or check in the last 5 minutes is skipped). A server that answers unauthorized is reconnected at once; only a clear sign-in signal counts (an HTTP 401, Copilot's own "OAuth challenge required", invalid_grant, an Entra sign-in code), so other errors restart nothing and a server that answers its tool list is not restarted by the check (a reconnect before a message after a quiet hour is a separate rule); a server is checked at most once every 5 minutes; one that does not answer in 30 seconds is marked Did not answer a check on its row (that request is not cancelled in the runtime, and no second one is sent while it is outstanding). Nothing shows in the chat while all is well, only Reconnecting … during a reconnect, and the agent gets no extra message from it. New Settings > Diagnostics (off by default): the activity-log row renewed (automatic) is now written only with it on. Probed on a local test server only; not checked on a real Entra server, macOS or Windows.
• The MCP sign-in renewal check no longer fails after the day it was written (spec 058, PR #1211's CI): shots-058-mcp-signin-refresh reads the Renewed 12:40 row against the run's own day, with the page clock held still, instead of a fixed date. Test driver only, no app change.
• Real office chat text is now checked against what is really behind it (spec 097, issue #1146): a new check measures every line of chat text, link, code chip and header text against the screen pixels under it, in the main and a helper chat at both window sizes, and fails under 4.5:1 or when header text runs past the window edge. The unreadable helper chat itself was fixed by #1164.
• Real office: the monitor screen shows who needs you (spec 097 FR-236, FR-238, owner decision): a desk's screen turns amber when its agent needs you (also on its empty desk while it waits at the Boss), red on an error, orange at the spending limit, and green when a finished agent's reply is not read yet, with a New reply tag above the name plate. The text on the screen stays readable on every colour. The corner call lights and the red and amber glows are gone. The arcade's cabinets do the same.
• One face per agent in Real office (spec 097, issue #1193): the left rail row showed a flat figure while the chat header showed the agent's Real office portrait; the rail, the header and the Needs you faces now use one picture. Checked in the simulator at 1440x900 and 1280x800 (check-097-one-face: Real office and Flat with two crews).
• Real office: rename a session from its name plate (spec 059, issue #1155): the same pencil as on an Open office desk appears beside the name when you point at or tab to a plate, and F2 on a focused plate now edits the name in place too (it did nothing in Real office before). Enter saves, Esc cancels.

Agent Durbar v0.1.10

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.10 (release, 2026-11-05)

• No empty More menu in a helper's chat (spec 005 FR-234, issue #1145): a sub-agent's chat header no longer shows a More button that opens an empty popover; a main agent's chat keeps More. Checked by the simulator driver check-helper-more at both sizes on Linux (not run on macOS or Windows).
• The release clip zooms on what changed (spec 008, owner ask): each scene of npm run release:video eases in to the part of the app that changed, holds, and eases back, with a caption for every step, so a viewer can follow it; -zoom stills let you check the frames. The privacy check now refuses any written-out home folder (/home/<name>, /Users/<name>, C:\Users\<name>), and the simulator's sample paths are neutral (~/projects/demo/...). Team tooling and simulator data, 0 AI credits. Checked by test:releasevideo; not run on macOS or Windows.
• Account limits per agent app (spec 011 FR-080..FR-094, issue #690): Settings > Agent apps shows what each provider itself reports about the account, separate from per-session usage: figures with their units, the source, when Durbar last asked and the reset time. Absent data says "not reported", never zero; a stale reading or a failed read is never shown as current. Read-only and free: Copilot account.getQuota (an account GitHub bills by usage reads "billed by usage, see GitHub", not a request count), Codex account/rateLimits/read; Claude Code has no free call and says "not reported". No polling. Near a reported limit (10 % left, awaiting owner review) it warns and names other ready apps; only your own click opens Continue in another app…. Both reads were checked on real Copilot and Codex accounts on Linux (2026-11-02): free, no session or turn started; Windows and macOS not checked.
• A very large file written by an agent no longer stalls Durbar (spec 012 FR-167, issue #1052): while a big Write streams in, the chat masks only the part it can show, instead of the whole file at every update. An 800,000-space write took about 9 s of CPU on Linux; on an Apple M4 it went from 2.2 s to 0.4 s. Checked with test:edits.
• Helper chat keeps its "Back to parent chat" bar in view (spec 015 FR-020, issue #1142): the "Sub-agent activity" bar stays at the top of a helper's chat while you scroll, on an opaque background, so the parent chat is always one click away. Checked by the simulator driver check-helper-bar at both sizes on Linux (not run on macOS or Windows).
• Agent rules shortened (spec 021 FR-063): two sentences in AGENTS.md say the same thing in fewer words (3,000 to 2,997); no rule changed. Checked with npm run rules:check and a before/after token diff.
• The merge gate refuses held PRs (spec 021 FR-079): scripts/pr-gate.sh now says "not ready" for a PR labelled status:blocked or needs-owner-approval, or one an owner has put on hold in a comment until an owner lifts it (only the date and link are printed). Checked with a fake gh and five planted mutants.
• The merge gate ignores quoted hold words (spec 021 FR-080, issue #1178): scripts/pr-gate.sh no longer treats on hold or do not merge inside code, a fenced block or a > quote in an owner comment as a hold (or a quoted hold lifted as a lift); a hold written as prose still refuses the PR. Checked by npm run test:trustedcontent (fake gh) on Linux; not run on Windows or macOS.
• Every world gets the features of Open office (spec 021 FR-078, owner decision 2026-11-04): a rule and a check, no change in the app. docs/feature-inventory.md now names an issue for each Real office gap (#1147 lists them: helpers, desk figures, marks, hover card, drag to rearrange, themes, characters, rename pencil, chat look), corrects cells that were stale, and the inventory check fails a world offered with a missing Open office feature. An owner decision waives only the cell it is written in and must read (owner decision, YYYY-MM-DD) or (owner decision, #NNN); a Real office works note with a limiting word (only, no, not, ...) counts as a gap, which gave the trouble signs #1171 and four chat-level rows #1172. Checked by npm run test:featureinventory, simulator-free.
• Test driver only: the helper-bar and helper-more UI checks look for the renamed read-only chat class (helper-ro), so main's UI checks pass again. No change for users.
• Needs you inbox: no giant avatars (spec 052, issue #1174): with Real office chosen (for example in the List view), the inbox could open with the waiting agents' faces several hundred px wide, hiding the requests and their Allow and Deny buttons. The faces are 30 px again. Checked by the simulator driver check-inbox-avatar at both sizes on Linux (not run on macOS or Windows, nor in the packaged app).
• Serious mode only hides the mood drawings; Motion is its own switch (spec 052 FR-069, issue #1182): Serious mode no longer stops walking, typing and lights, so the office does not look broken with it on. A new Settings switch, Motion (on by default), and the system's reduced motion setting stop the movement. Settings of an older version keep Serious mode as it was and have motion on. Checked in the simulator at 1440x900 and 1280x800 (check-serious-motion); not run on Windows or macOS.
• The rename pencil follows its name when a card's live line changes (spec 059, issue #1135): a new line of live activity could move the names in a room while their hover pencils stayed put, over the task line. Checked with shots-001-live-activity, now part of check:ui. The check runs once per animation frame and reads where the names are only when a live line changed height (check-office-ops: one layout read per live paint).
• A click on the middle of a short name opens the chat, not rename (spec 059 FR-030): the rename pencil is invisible until hover, and on a short name with no room beside it, its box could start left of the name's middle and take the click. The whole box now stays in the name's right half. Checked in the simulator at both window sizes.
• End-of-session summary on the lounge, in the briefing, with the agent app's own numbers (spec 085, issue #604): a finished agent resting in the lounge shows its files and lines in its bubble (click it: the changed-files list) and in the peek card's text; by keyboard the desk label names the numbers and the folder chip lists the files; Since you were away adds one line for the sessions you ended yesterday and today (numbers only, 7 days kept on this computer); and the card adds a second sentence with the totals GitHub Copilot reports (files, lines, AI credits, requests), named as reported and never merged with Durbar's own count. Claude Code's status line and Codex's notify are not used yet. Simulator, unit tests and the driver at 1440x900 and 1280x800; getMetrics was checked on a real GitHub Copilot session (it is free of model cost; a non-zero codeChanges was not seen), and the feature itself was not run end to end on a real session.
• A helper's chat in Real office no longer shows the floor picture behind its text (spec 097, #1140): every chat, main or helper, new or resumed, has the same world look and follows the Office layout setting, also when it changes with a chat open. Checked with the simulator driver check-097-chat-look at both sizes.
• No zoom under a still pointer (spec 097, #1040): pressing Esc back to the office with the pointer resting on a screen no longer opens that screen's zoom; the next real move does, and keyboard focus zoom is unchanged. Checked in the simulator at both sizes, with a mutant; not seen or verified on macOS.
• Status counts in the top bar keep their width (spec 097, Changes 2026-11-04): on a system font whose digits differ in width (seen on Windows), a count change made its pill a fraction of a pixel wider or narrower, which shifted the top bar and made it measure itself again. Each count now has a fixed width per digit. The simulator driver check-office-ops-real also sets motion on itself, so a Windows runner's "reduce motion" setting no longer makes its moving-people row find nothing, and it fails when any top-bar part changes size on a status update.
• Real office is lighter (spec 097 FR-180..FR-184, T149..T153, simulator only): a status change no longer re-measures the lobby's name chips or rebuilds name plates and folder signs (simulator op counts: layout reads per update 65 to 3 at 15 agents and 67 to 3 at 40, measured before the slips joined the room; with the slips 15, as they are drawn again on every update, T184).
• Real office's lamps no longer blend (spec 097 FR-180, T150): each lamp is a gradient tinted in advance, and an off lamp is a dark gradient. The look moves very slightly: a lit monitor is a little less amber. Light is the owner's call: the before and after pictures are in the pull request.
• Real office's lights move in steps (spec 097 FR-180, FR-181, T151): pulse and breath change about 3 times a second, and glows and call-light halos are static gradients instead of box-shadows. A screen's glow now starts right at its bezel.
• Real office's dialogs have no blur (spec 097 FR-182, T152): a dialog is the opaque world surface, its text checked at 4.5:1 or more on the computed colours; small menus and toasts keep the glass, ...

Agent Durbar v0.1.9

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.9 (release, 2026-11-04)

Highlights

• The agent rail: every agent beside the chat, by folder, and who needs you first.

All changes

• Still highlights (no motion) (spec 001 FR-192, issue #780, WCAG 2.2.2): Settings > Highlights has a switch that keeps the coloured glow and the New badge on desks that need you or have a new reply, but stops the movement that goes with them: the person's jumping and pulsing, the glow's breathing, the badge's pulse and the waiting badge's pings, the waiting person's waving hand and the tap on the desk after a minute. In the Real office the whole room stands still. It does not stop every animation (the system's reduced motion setting does that; some small ones, such as a theme's own figures, still follow it). Off by default and saved on this computer. A failed save says so and the switch shows what is really saved. durbar_settings (MCP) now reports stillHighlights next to seriousMode. Checked in check-unread (simulator, both sizes), test:stillhighlights (main's handler on a stand-in electron with a real store) and test:ipcchannels; not run on a packaged app, Windows or macOS.
• Long chats open faster, and keep following the end (spec 005 FR-210..FR-215): when you leave a chat, its older messages are marked with their measured height, so reopening a chat of 4,000 simulated messages lays out only the ones near the view (the timing of each run is printed by the driver, never asserted; a long history opened for the first time is not faster, nor is a chat closed from its Browser tab). The marks are dropped when the width or a layout setting (density, theme, font) changed (FR-213), and for a message that changed while the chat was closed (FR-215). A chat scrolled up, even by a little, is no longer pulled back to the end by a streaming turn, and a late scroll event no longer stops the chat from following a turn (FR-214). Scrolling, copying across messages, find and keyboard focus are unchanged. Simulator only; not run on Windows or macOS.
• Test driver only (issue #1102): the long-chat visibility check no longer fails on a loaded machine or on a 2x screen. No change for users.
• Test driver only (issue #1127): the long-chat visibility check runs one window size per CI slot, so it no longer hits the suite time limit. No change for users.
• The agent rail (spec 005 FR-190, FR-194, FR-216; spec 052 FR-215, FR-216): with a chat open, a rail on the left replaces the strip at the top. It lists every agent by folder, with the folder's count; whoever needs you comes first, in a taller row that says how long they have waited and what they ask, and their folder moves up. Your door at its top shows the Needs you count and opens the queue. A folder collapses and expands with a click, Enter, Space or a double click, still showing its count and how many in it need you; opening a chat in it expands it, and the collapsed folders are remembered per window. On a Mac the rail's top row and the chat header move the window. Checked in the simulator by check-rail at 1440x900 and 1280x800 (all 15 agents, no page overflow, two row heights, contrast) and test-railorder; not tried on a real Mac or Windows.
• A short clip of a release's changes, as a trial (spec 008 FR-171..FR-178, owner request): npm run release:video -- --version <tag> --out <dir> records 30-60 s of the simulator (title, one captioned scene per chosen change, end card) with the captions in a band under the app, and writes an MP4 and a GIF (ffmpeg needed), refusing to finish when a scene's text shows a local path, the user or host name or a token shape. A fragment may now end with optional Highlight: and Scene: lines (changelog.d/README.md), which the fold drops. The release notes now start with a Highlights block built from those lines, and the release steps in changelog.d/README.md make the clip part of the flow, starting with v0.1.9 (the first clip is shown to the owner before it is published; older releases get none). Team tooling, no app change, 0 AI credits. Checked by test:releasevideo and a v0.1.7 trial on Linux; not run on macOS or Windows.
• Open with menu in the chat header (spec 009 FR-019..FR-027, #685). The header's Open in VS Code button is now Open with: it lists the installed VS Code, VS Code Insiders and VSCodium, and Show in folder for the system file manager. It opens the agent's folder only and says the conversation stays in Durbar; Continue in another app… (More menu, when a second agent app is installed) is still the way to hand it over. A folder that is gone or a file is refused with a message; a failed launch says so. Something the file manager would run is selected, not opened. A project folder that is a symbolic link is opened at its real path, with a note saying so; a link to a file or to nowhere is refused. The folder status dialog's Open in editor and Continue in VS Code go through the same folder checks and use the first installed edition that can open the folder (on Windows, the next one when the first is only a .cmd and the folder's name has a character cmd.exe could misread); the profile editor's button is renamed Open in editor but opens the profile file as before. Simulator, unit tests and a driver at 1440x900 and 1280x800; no editor or file manager was launched by a test, and the one real check (a folder opened in Finder) ran before the later changes.
• Agent events are checked in the app (spec 010 FR-035, issue #1039): every event from Copilot, Claude Code or Codex now passes the same shape check the tests use before the window or any other part of Durbar sees it. A malformed one is dropped, with a log line that names only the event type and the fields at fault, never their values. Checked with test:runtime-core and one short real session per app (149 events, none dropped).
• Settings: "Saved" only when it was saved (spec 011, issues #1094, #1047): asking again for a value already in force while the settings file cannot be written no longer says Saved, and a failed save no longer leaves a .tmp file next to the settings. Checked by test:credits (injected write and rename failures, mutation-checked).
• Test driver only (issue #1136): the Codex stop-escalation test waits for the signals instead of a fixed sleep, so it no longer fails on a loaded machine. No change for users.
• No MCP config left in the temp folder (spec 012, issue #1093): the private folder holding mcp-config.json (MCP env and header values) for a Claude session is now removed also when the app quits while Claude still runs; every other ending (spawn failure, error, kill, close) was already covered and now has a test. npm test no longer leaves durbar-* folders behind and fails a suite that does.
• A helper started in the background no longer reads Done while it works (spec 015 FR-018, issue #1141): when an agent launches a helper in the background (Claude Code's run_in_background, Copilot's background task), the agent's own call returns at once; the helper's chat block, card and transcript now stay Running until the helper itself finishes or fails, and a helper still running when its session stops reads Failed ("Stopped with its session"). Foreground helpers are unchanged. Checked with mapper tests and a simulator driver; not seen on a real session.
• Auto-follow work, opt in (spec 052 FR-182..FR-190, issue 686): a new switch in Settings (off by default, kept on this computer; main answers a real ok) makes a sent message take you back to the office with that agent's desk marked and focused, and open its chat (focus on the chat title, never on Allow or Deny) when it finishes, asks, errors or reaches its limit. It never answers or approves anything, waits while you type, have a dialog or menu open, or the window is in the background, stops following when you open another agent yourself, and respects reduced motion (no animation). A strip under the top bar shows Previous and Next over the attention queue (the Needs you inbox order, then unread replies) with the position and count (keys [ and ]), a Pause button, and, when nothing needs you, Next returns to the office with the last desk focused. Checked in the simulator (check:ui check-auto-follow, both sizes, 15 agents) and test:autofollow; not run against a real agent session.
• Needs you text no longer runs words together (spec 052 FR-214): the inbox, the answer bar, the announcements and the Real office slips read "search_companies) Allow for this session ...", not "search_companies)Allow ...", and a card with several questions shows all of them. Checked by the simulator driver check-inbox-text.
• MCP sign-ins kept fresh (spec 058 FR-204..FR-217, first part): before a chat's first message after an hour without a reply (or a wake from sleep), Agent Durbar asks Copilot or Claude Code to reconnect the chat's remote MCP servers, so the app renews an expired sign-in with its own refresh token. The message waits at most 5 s under one quiet line; Stop sends nothing; a server that still needs a sign-in gets a line with Sign in…. On by default (Settings > MCP servers); in Copilot's sandboxed modes (Read-only, Ask first) only the sign-in line, since a reconnect there did not renew the sign-in (Linux probe); Durbar never sees a token; n...

Agent Durbar v0.1.8

Note

Corrective patch release for #1084: Copilot MCP sign-in for Microsoft Entra servers works again. The bundled Copilot SDK is pinned back to 1.0.24 as a stop-gap while a lasting route is built; newer SDKs need the host app to bring its own Entra app, which Agent Durbar does not do.

Unsigned build of Agent Durbar. Not code-signed or notarized: see the README for the macOS Gatekeeper and Windows SmartScreen steps. Needs at least one agent app signed in: GitHub Copilot, Claude Code or OpenAI Codex (Settings > Agent apps can install it). Independent project, not affiliated with GitHub, Microsoft, Anthropic or OpenAI.

v0.1.8 (release, 2026-11-03)

• Settings that cannot be saved are said so (spec 011, issue #1023): when the settings file cannot be written, a switch (serious mode, notifications, break resume, update check, chat links, default cap, characters, places) no longer reports success and then forgets the choice at restart: it stays as it was and the row shows "Could not save this setting. It stays on." Things that must keep working stay in force for the current run: the session browser's first-use note, an update announced once, the last new-folder location, the theme, turning the OS-sandbox always-continue choice off and a lower default cap; where it matters the window says it may not last after a restart. Checked in test:credits, test:sandbox-ops, test:browseripc and the simulator driver check-settings-save (both sizes); no Windows or macOS run.
• Codex no longer starts a second copy of each local MCP server on every status read (spec 012 FR-164, issue #1077): the MCP status at session start, with the first message, after a sign-in challenge and in the context figures was read with one request that made Codex 0.163.0 run every local server's command again (in Read-only too). Each server is now read by name on the session's own connection; the names of your own config.toml servers come from Codex's config/read, which starts nothing. Checked by unit tests and by driving the built adapter against the real codex app-server with no prompt (a start marker per stdio server stayed at 1 over a start, three status reads and two context reads; 7 before). Not run on macOS or Windows or with a signed-in account.
• Activity export is written safely (spec 052, issue #973): Tools > Activity > Export now writes through a temp file and a rename, mode 0600, and refuses a link, hard link, folder, FIFO or device at the chosen name instead of writing through it, like the notes export. Content and dialog are unchanged. Checked with real temp folders in test:activity (link target untouched); not run on Windows or macOS.
• Text popups stay open while you select their text (spec 052 FR-237): a drag that starts in a dialog and ends outside it (selecting a field's text and overshooting) no longer closes it. It did for 11 popups: the Rename, Rename floor, Place, palette, Schedules, spending limit, New session and New profile dialogs, the floors popover's new floor name, the Needs you inbox and the keyboard sheet. A click into a folder's rename field no longer closes the Folders panel. One shared helper decides; checked in the simulator with real mouse and keyboard input at both window sizes, and by unit tests.
• MCP servers: the chat's actions in Settings, and a health check at launch (spec 058 FR-195..FR-203): Settings > MCP servers now has Sign in and Sign in again… on a chat's row (they open that chat's MCP list and start the same sign-in there) and Restart where it failed (N) for a server that failed in several chats; a server of your agent apps' lists whose command cannot be found says so. When the app starts, one calm strip lists the MCP servers that need attention, with the reason and a suggested fix for each; nothing is fixed, started or sent for you. Simulator and unit tests only; not checked on a real agent app, macOS or Windows.
• Fixed: Copilot MCP Sign in for Microsoft Entra servers (issue #1084): since v0.1.7 it failed with "Microsoft Entra authentication requires the host to configure an Entra client ID", because a dependency update brought a Copilot runtime that needs an app registration Agent Durbar does not have. The Copilot runtime is back to the one from v0.1.6 (SDK 1.0.24). Not checked with a real Entra sign-in yet.
• MCP dialog: a visible Tools button, tool rows that fit, and tokens for the ticked tools (spec 069 FR-067, FR-068): each server row has a Tools button (wrench, "Tools (24)" or "Tools: 20 of 24 on") instead of only the tiny arrow; long tool descriptions stay inside the card (2 lines, full text on hover); each tool shows "about N tokens" and the server's token figure counts only the ticked tools until Restart session. Simulator checks (shots-069-mcp-tools, both sizes, two looks, three mutations); not verified on macOS or Windows.
• Settings > Agent apps: a Homebrew app with an update waiting no longer shows "Could not check: exit 1" (brew outdated exits 1 when it finds one). A failed check now also shows a short, masked detail line. (#1081)
• Copilot's sandbox switch checked on macOS (spec 087 K16, issue 834): a real probe shows the OS sandbox clears in place on a switch to Allow all and comes back on the way out, as on Linux. The probe's own cleanup now handles what macOS leaves behind. No change in the app.