Skip to content

0.9.2 — multi-instance asq, plus fixes and optimizations across the fleet

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 03 Nov 21:06
6a77e65

Run asq at your Mac and over SSH into the same Mac at once, on one home and one fleet: each asq shows the others, the one over SSH shows the Mac's usage, and an asq of another release won't start beside a running one. Around that come fixes and optimizations across the fleet. Pause a project's agents, or every project's, so idle agents stop spending tokens, and resume each later on its own session, claim and worktree, even after a reboot. Spawns follow the machine's load instead of a fixed four per project. asq asks fewer questions and asks them in place, buttons work as soon as you see them, and a double-click never confirms the dialog it opens. aisquare outcomes says what your agents got done this week, and aisquare fleet top what each one consumes. An idle account whose token went stale is refreshed quietly instead of reading as signed out. Under the hood, tests that failed on Windows or now and then are fixed, and Dependabot's four version bumps are in.

0.9.2 builds on 0.9.1. It gathers these pull requests:

  • #472: several asqs on one home. It combines six pieces: #461 (usage over SSH), #462 (the instance gate), #463 (the pane size), #464 (the instance view and the SSH tmux guard), #465 (test dates) and #466 (the look for an older asq in ps).
  • #457: pause and resume.
  • #485: five items, each from its own branch: the quiet refresh of a stale login, real repository names and the sidebar's live count, aisquare outcomes, auto admission, and aisquare fleet top.
  • #491: the second release review's fixes in #485's code: auto admission never holds a manager or a resume, and max_agents_per_project takes auto or a number of at least 1.
  • #489: fewer clicks, from three pull requests: Stop's Don't ask again (#477), yes/no questions asked in place (#486) and a click guard that waits only until the screen has redrawn (#478).
  • #495: the third release review's fixes: a double-click never confirms the dialog its first click opened, Unset in a role binding's Unset key… dialog is guarded too, and a test flake.
  • #455: rename an agent.
  • #456: a project's rename reaches its codename and tmux session.
  • #458: one sidebar highlight.
  • #475: three asq features, each from its own pull request: the faded side (#470), the active-only filter (#467) and the Onboard view's ↻ (#468).
  • #454: Usage & stats and the Accounts table.
  • #469: TahirRabia's fixes, #447 to #450.
  • #479: the opt-out notice.
  • #494: CodeQL alert #23 in aisquare outcomes.
  • #488: Dependabot's #481 to #484 (setup-uv, certifi, ruff and truststore), with a Windows test fix.
  • #459: a flaky test.
  • #476 and #480: Windows CI for #472's tests, and the review fixes for #469 and #472.

#455, #456 and #457 each went through several rounds of review and testing. #457 was also run against 0.9.1's own code on a store it had paused, and a UI tester checked #454 on screen at 160x48. Each of #461 to #466 was tested and reviewed on its own before #472 combined them. #472's SSH guard rests on a probe from a real SSH login into macOS 27.2. #469 brought in TahirRabia's four fixes after three review rounds and a tester's round. Each of #485's five items was tested on its own, and #485 was approved in its fifth review round. #467, #468 and #470 were each tested on their own and validated together before #475 combined them. #477, #478 and #486 were each reviewed and tested by hand in a real terminal, with real mouse clicks, before #489 combined them. #495's double-click tests fail without its fix and pass with it. #488 and #495 each ran the full CI matrix, Windows included, before they merged. The release ran the full CI matrix green: the test suite on Linux (Python 3.11 to 3.13, with the captain on and off) and on Windows in 10 shards, and the installer checks on Linux, Windows and macOS.

asq at the Mac and over SSH, at once

You can now run asq at your Mac and over SSH into that same Mac at the same time, on one home and one fleet.

  • Each asq sees the others. While two or more are open, a line under the sidebar's status line says how many there are and where each runs: 2 asq · Mac + ssh (here). Hover it to see each one's version, host, pid and terminal. aisquare instances lists them (--json too), and the doctor's asq instances row warns when one runs another release.
  • One release per home. An asq won't start beside one of another release. It names the other asq and asks Quit the other asq? [y/N]. On y, the other quits as if you had pressed q in it, and its terminal says which asq asked. On n, the new one starts nothing and tells you how to quit the other yourself. It checks before it opens the store or touches tmux, so an asq that is refused has changed nothing.
  • A 0.9.1 asq is found too, and never killed. 0.9.1 writes no record, so the new asq looks for one in ps. It names that asq's pid, terminal, start time and command, and says what kill <pid> would lose compared with quitting it with q.
  • Usage over SSH shows what the Mac recorded. An SSH login can't read the login Keychain, so in 0.9.1 its asq showed no usage. Now it shows the readings the Mac's asq took for the same login, never another login's, marked · Mac and via the Mac's asq · 2m ago. When the Mac sees a sign-out or an expired login, the SSH side shows it too.
  • A shared pane takes the size of the asq you last typed in. When both asqs show one agent, a key, a paste or a click makes that asq size the pane. The other asq shows [sized by the other asq (ssh)] instead of resizing it back. In 0.9.1 the two kept undoing each other's size.
  • SSH doesn't start the fleet's tmux server on macOS. Agents in a server started from an SSH login can't read the Keychain, so a start from SSH is refused with start asq on the Mac first. Once the Mac has started the server, agents spawn into it from SSH as usual.
  • Upgrade and uninstall wait for the other asqs. aisquare upgrade and aisquare uninstall refuse while an asq of 0.9.2 or later runs on the home, and asq's own Update and Uninstall refuse while another one runs. Each refusal names the asqs to quit first.

Running more than one asq has the details.

Pause the fleet, and pick up where it stopped

You can now stop your agents spending tokens without losing their place, and bring them back when you are ready.

  • Pause one project, a group or all of them. aisquare fleet pause exits each running agent's Claude process as fleet stop does, the manager first, and keeps its label, task claim, worktree, branch, account and transcript. -P <project>, --group <group> and the new --all and --managers choose which agents. The captain, and an agent that runs the command, keep running.
  • Resume brings back exactly the agents a pause stopped. aisquare fleet resume starts each one under its own label, resumes its session from the transcript (claude --resume) and types one carry-on line, with the manager last. An agent that exited without a pause stays 💤, and running resume twice does nothing the second time.
  • Nothing is handed out while they are paused. A paused agent keeps its claims however long the pause lasts. The board says agent_paused, which wakes nobody, instead of agent_exited.
  • Pause before a reboot or an OS update. A paused agent outlives its tmux server: fleet reap --server-down leaves it paused with its worktree and claims intact. reap, Restart all, Stop all and fleet shutdown leave it alone, and a new spawn takes the next label instead of its label.
  • Pause and resume from asq. ⚙ Settings ▸ Fleet has Pause all… and Resume all… beside Restart all… and Stop all…. A project card's ⋯ menu and its Settings tab have Pause agents… and Resume agents…. Each opens one dialog that lists, by project, the agents it will act on. A paused agent shows ⏯ paused in the sidebar and in fleet ls. When a project's manager is paused, its Manager tab offers ▶ Resume agents….
  • Resume one agent, or the whole project. A paused agent's view offers Resume and Stop. Resume brings back that agent alone, as fleet restart <label> does, and the project stays fleet-paused until Resume agents… or fleet resume. Stop ends the pause for good.
  • Only you end a pause. You end one from asq, Remote or your own shell, with no --as. When an agent tries to stop, restart, switch or resume a paused teammate, the command is refused whatever --as it passes, and the captain acts as you only from its own window.
  • A paused agent whose task closed is not woken for nothing. Your resume ends its pause instead, returns its claims to the board and says so. A paused manager whose task closed stays paused until your fleet stop manager.

Spawns that follow the machine's load

By default, 0.9.2 no longer stops each project at four agents. It decides by how busy the machine is.

  • max_agents_per_project now defaults to auto. A spawn is admitted unless the machine is busy. While it is busy, a spawn into a project that already runs 3 live agents is held, and a project with fewer is always admitted.
  • Busy means any one of these: a 5-minute load average of 1.0 per logical CPU; on Linux, CPU pressure (some avg60) at 40; memory pressure at warn; on macOS, thermal pressure at heavy. Each threshold, the floor of 3 and the hold are [fleet] busy_* settings.
  • A fan-out is paced even on a calm machine. The load average moves minutes after a burst of spawns does, so past the floor a project gets at most 2 new agents every 5 minutes. Ten spawns started at once into an empty project leave 5 running, and the rest are refused as held.
  • A held spawn never waits. It is refused at once with capacity_busy, and the message names the signal, its value, its threshold and the earliest retry. While aisquare watch runs, it wakes the project's manager when capacity frees up, so the manager retries without polling.
  • The manager, the captain and a resume are never held. However busy the machine reads, a project's manager is spawned and restarted, and a resume brings back every agent the pause stopped. The manager still counts toward the floor of 3 for the agents after it. A new coder or reviewer is held as before.
  • See why. aisquare fleet capacity and the doctor's fleet capacity row list each signal with its value, threshold and verdict, and the holds in force.
  • The hard limits still apply, to the manager and a resume too: min_free_mb, critical memory pressure, max_agents_total and a group's max_agents. A restart or switch of a running agent is never held.
  • A number of 1 or more keeps 0.9.1's behaviour exactly. To keep the old cap, run aisquare config set fleet.max_agents_per_project 4.

Fewer clicks in asq

asq now asks fewer questions, and asks them where you clicked.

  • Stop without the dialog, if you choose. The Stop dialog has a Don't ask again tick. Tick it and press Stop, and from then on the agent view's Stop and the sidebar's x stop an agent at once. They stop it gracefully, as the dialog's Stop does (never Force), for every agent, the manager and the captain included. A toast says stopping <label> … until the stop answers. ⚙ Settings ▸ App settings ▸ User preferences ▸ Advanced ▸ Ask before stopping an agent turns the dialog back on, and so does aisquare config set ui.confirm_stop true. Until you tick it, Stop asks as it did in 0.9.1.
  • Five yes/no questions are asked in place. Remove from fleet (a card's ⋯ menu), a role binding's Clear, Reset all settings to default, Take over (Spawn agent, handing off from a teammate) and attaching a persona to a running agent no longer open a dialog. The first click, or Enter, turns the button into its question (Remove? click again) and says beside it what the second will do. A second click, a double-click or Enter again goes ahead, and Esc or three seconds put it back. Stop (until you tick Don't ask again), Start fresh, Discard, Restart all…, Pause all…, the update and the usage-sharing consent keep their dialogs.
  • A button works as soon as you can see it. In 0.9.1, Stop and Restart in an agent's view, the bulk dialogs' Confirm, Start fresh and a few others ignored every click for a second after the view changed, with a toast to say so. Now they ignore only a click that reaches asq before the redrawn screen has been read off it (by your terminal, or by tmux or sshd in between): the click aimed at what was there before. When that reader can't keep up, a click is held a second at most, and no toast is raised.
  • A double-click never confirms the dialog it opens. With a small fleet, the dialog of Stop all… or Pause all… can draw its Confirm right where the button that opened it was. Each of those guarded buttons now ignores the second click of a double-click unless it took the first too, so the dialog stays up, and Enter then cancels it. A click of its own on Confirm still acts as soon as Confirm is drawn. Unset, in a role binding's Unset key… dialog, is guarded the same way now: with three bindings, the page could be scrolled to put a binding's Unset key… under it. A double-click slower than half a second, or whose two clicks land on different cells, still counts as two clicks.

What your agents got done: aisquare outcomes

aisquare outcomes is a weekly summary of what your agents achieved and how much of it you steered.

  • One week, on your local clock: Monday 00:00 to now, against the week before up to the same point. --week last or --week 2026-W43 picks another week, and --json gives the same figures.
  • Outcomes, per repository by its real name: the pull requests you authored that were merged and opened, your commits and their lines in the checkouts the agents worked in, and board tasks done and reopened. The pull requests come from one GraphQL call through gh, the report's only network call. Without gh it skips them and says so.
  • Effort: agent-hours, sessions, turns, tool calls, files edited, tokens, list-price cost, and the cost per merged pull request and per done task.
  • Who steered: the share of turns a person started against those an agent started, agent-minutes and tool calls per human prompt, interrupts and denied tool uses, and each session as directive, guided or collaborative.
  • Every turn now records who started it. A prompt you type is human:typed, a tell from your terminal, asq or the phone is human:tell, and a manager's tell is agent:tell. A wake, or a Stop hook that keeps a turn going, is an agent's. Turns from before 0.9.2 are inferred by the same rule, and the report says which are.
  • The first run reads the transcripts once. asq's usage sweep keeps a cache of them filled, so a weekly run takes a few seconds. Without that cache, the first run reads a week or two of transcripts first: 17 s for 2 GB of them on the machine it was measured on.

aisquare outcomes --help says how each figure is measured.

What each agent consumes: aisquare fleet top

You can now see which of your agents is using the machine.

  • One row per live agent of every project. Each row has the CPU, resident memory and number of processes of the agent's whole process tree (claude and what it runs: MCP servers, a pytest, a build), how long it has run, and its session's tokens and list-price cost so far. The heaviest CPU comes first, and a totals row ends the table. The line above the table counts the live agents as the sidebar's header does, with the captain named apart (3 live agents + the captain).
  • From a shell or in asq. aisquare fleet top prints the table once, with --sort cpu|mem|tokens, --project and --json. asq's ⚙ Settings ▸ Agent resources page shows the same table and reads again every 5 seconds while it is open. Keys 1 to 3 sort it, and r reads at once.
  • Light, and read-only. Each read is one ps -A and one tmux listing for every agent at once, about a quarter of a second for 26 agents. Nothing on the page stops or restarts anything. CPU is ps's figure, in percent of one core; on macOS it is an average over about the last minute.

Rename an agent or a project, and the name follows

You can now call an agent or a project what you like, and use that name everywhere.

  • Rename a running agent from its sidebar row. Double-click its row, or highlight it and press F2, then type any name. The line under the field shows the label it will become (Login Fix → login-fix). From a shell, aisquare fleet relabel coder-1 login fix does the same and prints ✓ coder-1 is now login-fix.
  • The new label works everywhere at once. fleet tell, stop, restart and switch, the board's lines and the agent's tmux window all use it. The old label finds nothing until a later spawn is given it, and the refusal says what the agent is called now. An agent_renamed line tells the manager on its next turn.
  • A taken name gets a suffix, and you are told. If another running or paused agent holds the name, yours becomes login-fix-2. The manager and the captain keep their labels, and no agent can take manager or captain.
  • Renaming a project renames its codename and tmux session. Save a card's name, the Settings tab's nickname or aisquare project rename "AISquare CLI", and the codename becomes aisquare-cli. The tmux session asq-earnest-hawk becomes asq-aisquare-cli, and open panes stay attached.
  • The old codename keeps working, and stays yours. --project earnest-hawk still finds the project, and no other project is given that codename while this one is registered. Each board the project's agents read, a paused agent's included, gets one note, so a live manager learns the new name on its next turn.
  • You can see the codename your agents use. When a card's name doesn't give its codename, the card shows the codename dim under the name (· earnest-hawk). The name's tooltip shows the codename, its tmux session and --project.
  • Choose a codename yourself. aisquare fleet rename now takes names like billing-v2. It refuses a name that --project already uses for another project, such as that project's folder name or a codename it gave up. An exact codename in --project now wins over another checkout's folder name, where 0.9.1 refused it as ambiguous.

The sidebar: one highlight, a live count, and only the live projects

You can now move through the sidebar with the mouse and the arrow keys and always see exactly one highlighted row. Its header tells you how many agents are live and can hide the projects that have none, and asq shows which side has the keyboard.

  • The header counts the live agents: Fleet · 19 live. It counts every live agent of every project except the captain, including those in a folded group, a hidden directory or a forgotten project. With none live, it reads Fleet.
  • List only the projects with a live agent. Click the ○ after the header's + (it turns ●), or press l while the sidebar has the keyboard. The sidebar then lists only the projects with an agent working, waiting, asking for you, limited or unknown, and projects come and go as their agents start and exit. A paused agent doesn't count, so a project you paused leaves the list until you resume it or turn the filter off. A group header or the 📌 Pinned label with nothing left to show goes too, while the Captain section always stays. asq remembers the choice when it restarts.
  • The side without the keyboard fades. With the keyboard in the sidebar, the agent's pane or the page beside it is drawn slightly fainter, and the other way round. In 0.9.1 only the one-column divider showed where your typing would go, so keys meant for an agent could land on the sidebar's one-key shortcuts.
  • A click and the arrows move the same highlight. Click a row, press ↓, and you go on from the row you clicked. Enter, F2, x and the other sidebar keys act on the highlighted row, and x asks before it stops an agent, unless you ticked Stop's Don't ask again.
  • The arrows open what they land on. ↑ and ↓ open a project, an agent or Settings as a click would, and keep the keyboard in the sidebar, so the next ↓ never types into Claude Code. Enter hands the keyboard to the agent's pane.
  • Rows that act wait for Enter. Passing over +, Spawn agent, Start captain, the ⬆ update row or a group header only highlights it, and Enter or a click acts. Spawn agent reached by the arrows opens on the project whose view is on screen.
  • Holding an arrow doesn't open every view on the way. You end on the row you stopped at.
  • The Onboard view shows a repo you just cloned. In 0.9.1 its folder tree read the disk once, so a repo cloned after the view opened could only be onboarded by typing its path. Now ↻ at the start of its hint line, or ctrl+r anywhere in the view, reads the folders again, with open folders still open and the cursor on its row. Opening the view again with + reloads it too, and a path you typed before its folder existed turns ✓.

Fixes

These are the screens 0.9.1 cut short, and the things it got wrong.

  • An idle account no longer reads as signed out because its token went stale. Claude Code renews a login's token only during a session, so an account no agent had run on lately showed Log in, though the login was fine. Now aisquare has Claude Code refresh the token quietly (no window, no keystroke, no model request) and reads again. Only if that fails twice, a minute apart, does it say what it said before. The Accounts page, aisquare accounts usage, accounts list --usage, doctor --live, the status line, the sidebar and [accounts] before_limit's check of an agent's own account all refresh a stale token this way, one refresh at a time per account across every asq and aisquare process. A spawn's or a launch's account pick never waits for the refresh.
  • Stats name each repository by its real name. The Stats page's Top repositories and aisquare metrics stats show the git remote's repository name, spelled as its URL spells it, else the checkout folder's name. In 0.9.1 they could show a project's nickname, and you couldn't tell which repository was meant. Nicknames stay on the sidebar's cards and in fleet ls.
  • Settings ▸ Usage & stats scrolls to its last section. In 0.9.1, at 160x48, you could not reach Accounts, Busiest days, Models or Features.
  • The Accounts table fits a 160-column terminal beside the default sidebar, with no sideways scroll. The resets columns, now headed 5h resets and wk resets as in Claude Code's status line, show the distance alone (in 4h 59m, in 6d 12h), with the whole reset on hover. A long name or email ends in … and shows in full on hover.
  • The Fleet line says "1 account". On the Accounts page and in aisquare accounts usage it said 0.8 of 1 accounts left.
  • aisquare task add adds a done or dropped task's title again. In 0.9.1 the closed task kept the title's key, so the add answered already tracked forever, and recurring work such as a release checklist could not be added again. It now gets a new number and says (supersedes dropped #4), and so do the MCP task_add tool and the captain. A title whose task is still open stays idempotent.
  • The prompt log records your prompts only. aisquare log no longer lists Claude Code's own blocks as your turns: a background task's notification, a slash command's echo, a hook's reminder. A prompt that mixes them with your words keeps the words.
  • A leading --json gets usage errors as JSON. aisquare --json nosuchcmd and aisquare --json --badopt now print one JSON error object and exit 2, as --json's help promised.
  • aisquare project fresh --help reads as one paragraph, and --dry-run works from a fleet agent. A manager can read the reset plan from its own pane now. The real reset is still refused there.
  • Turning usage sharing off while its first upload is out no longer loses the opt-out notice. In 0.9.1, if that upload's reply landed while the switch was being written, no notice went, and the server went on counting the install as opted in. Even then, nothing was sent after sharing was off, and nothing was kept. The one notice is sent, best effort, as before.

Under the hood

These changes are internal. A test that counted accounts list rows by their first word failed at some temp-path lengths. It now counts a row only where one starts (#459). Three tests were pinned to fixed dates that a 7-day or 31-day window would outrun, and the first of them turned CI red on 2026-10-30. They now read the clock the code reads, and pass 400 days ahead (#457; #465 made the same fix, and #472 keeps its history). Six of #472's tests asked Windows for what only POSIX has, and a task-numbers test raced. All of them pass on Windows now (#476). A load-probe test expected a load average on Windows, where asq reads none, and it now runs both branches on every OS (#491). On Windows two asq records can share a start time, so a test that expected them in start order now sorts by id too (#488). A usage-sharing test failed about once in 700 runs, when the install id it prints had 404 in its first 8 digits, and it pins the id now (#495). aisquare outcomes now compares a remote's host with github.com whole before it asks GitHub about the repository's pull requests, which clears CodeQL alert #23; no real remote's behaviour changes (#494). Dependabot's four bumps are in (#488): certifi 2026.10.23 and truststore 0.10.5 in uv.lock and CI's constraints, ruff 0.17.3 for lint, and astral-sh/setup-uv 10.3.3 in CI. pyproject.toml is unchanged, so an install resolves its dependencies as before.

Installing, and pinning this release

The one-liner is the same as before, and installs the latest release:

  • curl -fsSL https://raw.githubusercontent.com/AISquare-Studio/aisquare-cli/main/install.sh | sh
  • on Windows: irm https://raw.githubusercontent.com/AISquare-Studio/aisquare-cli/main/install.ps1 | iex

Pin this release with its own installers, attached below. Every release from 0.9.0 on carries an install.sh and an install.ps1 that install exactly that version:

  • curl -fsSL https://github.com/AISquare-Studio/aisquare-cli/releases/download/v0.9.2/install.sh | sh
  • on Windows: irm https://github.com/AISquare-Studio/aisquare-cli/releases/download/v0.9.2/install.ps1 | iex

Already on 0.9.1? asq offers ⬆ 0.9.2 · restart to update under ⚙ Settings, and Restart & update keeps your fleet running. Or run aisquare upgrade. Quit any other asq first, at the Mac and over SSH: the 0.9.2 asq won't start beside a 0.9.1 one. --version V still overrides a script's own version, and --version latest drops it.

Upgrading:

  • Quit every other asq before you update. A 0.9.1 asq left open, say over SSH while the Mac's asq takes the update, keeps the 0.9.2 asq from starting. The 0.9.2 asq names it (pid, terminal, start time and command) and exits 1. Quit it with q in its own terminal, or end it with kill <pid>, then start asq again. Nothing is killed for you. From 0.9.2 on, an asq of another release is asked to quit instead.
  • The store migrates from schema v40 to v43 automatically on first open, in three steps, and nothing is rewritten or dropped. v41 adds a table for the codenames a renamed project gave up, v42 adds one nullable column, fleet_agent.paused_at, and v43 adds another, metric.origin (who started each turn). Running several asqs adds no migration. If you are coming from 0.9.0 or older, read 0.9.1's upgrade notes too.
  • [fleet] max_agents_per_project now defaults to auto (it was 4). It is the one default that moves, so you get it without changing anything. A number set in your config file is kept. A 4 you chose in 0.9.1 matched the default, so it was never written to the file. To keep the old cap, set it again with aisquare config set fleet.max_agents_per_project 4, or in App settings.
  • max_agents_per_project takes auto or a number of at least 1. 0.9.1 also took 0 (which refused every spawn), a negative number and true (a cap of 1). In 0.9.2 a config file with one of them is invalid, as aisquare doctor says, and config set can't change it. Set a number of 1 or more before you upgrade, or take the line out of ~/.aisquare/config.toml by hand afterwards, which leaves auto.
  • Moving back to 0.9.1 with aisquare upgrade --version 0.9.1 is safe. Quit every asq first: the command refuses while one runs. 0.9.1 opens a v43 store as it is. A paused agent is missing from its sidebar and fleet ls, fleet ls --all lists it as 💤 exited(0), and Restart all leaves it. --project <old codename> finds nothing there, so use the new one. 0.9.1 ignores what several asqs leave on a home: the ~/.aisquare/instances/ folder, the usage-login: rows in ui_state, and the @asq-size-owner option on tmux windows. It ignores the sidebar's active-only choice too, so its sidebar lists every project, and back on 0.9.2 the filter is as you left it. It ignores [ui] confirm_stop, so its Stop asks again, and a save there keeps the line for 0.9.2. It ignores the busy_* settings, and its default is 4 again. But it can't load a config file where you typed max_agents_per_project = "auto" by hand, so take that line out first. A config save in 0.9.1 (a config set, or a setting changed in its asq) drops a max_agents_per_project = 4 from the file, since 4 is 0.9.1's default, and keeps the busy_* settings. Moved forward again, the setting is auto, so set 4 again.
  • Resume paused agents before you move back. On 0.9.1 a paused agent's task stays claimed until fleet restart <label> resumes its transcript or you run aisquare task release <id>. A spawn there can take its label, and reap removes its worktree once its work is clean and pushed (the branch stays until the default branch has its work). Back on 0.9.2, a paused agent whose label another row took is not resumed.
  • Settings are added and none removed. No environment variable, extra or dependency is added or removed. [fleet] gains busy_floor, busy_burst, busy_load_per_cpu, busy_cpu_psi, busy_memory, busy_thermal and busy_hold_minutes. There are four new commands, aisquare fleet relabel, aisquare instances, aisquare outcomes and aisquare fleet top, and two new flags on fleet pause and fleet resume, --all and --managers. [ui] gains confirm_stop, on by default, so Stop asks first until you tick Don't ask again. The sidebar's active-only filter is no setting: asq keeps the choice in its store, and config set doesn't take it.
  • aisquare upgrade and aisquare uninstall refuse while an asq runs on the home (asqs of 0.9.2 or later; quit a 0.9.1 asq yourself). 0.9.1 went ahead. A script that runs upgrade --yes beside an open asq now exits 1. Quit the asqs first.
  • fleet pause now exits the agents, not only the signal. In 0.9.1 it set the project's fleet-paused signal and nothing else. Now it also pauses every running agent, and fleet resume starts them again. If a script ran fleet pause just to hold a manager's spawns, run aisquare team signal fleet-paused on in the project instead (it takes no -P), and off to clear it. An agent's fleet resume no longer clears a signal-only pause: it is refused and exits 1.
  • Only the owner ends a pause. Anyone may pause, and a manager that pauses its team keeps running. No agent can stop, restart, switch or resume a paused teammate. Your fleet stop <label> ends a pause for good and returns its claims to the board. Your fleet restart <label> resumes that agent alone, and the project stays fleet-paused (so its manager spawns nothing) until fleet resume or Resume agents….
  • A project rename now renames its tmux session. A script that attaches to asq-<codename> by name needs the new name, while aisquare fleet attach finds it by itself. Branches (fleet/<old codename>/…) and worktrees keep their names.
  • Projects keep their codename until you rename them. A project you nicknamed before 0.9.2 keeps its generated codename, which now shows dim on its card. Run aisquare project rename "<its name>" with the same name to have the codename follow it.
  • The sidebar's arrows open rows now. ↑ and ↓ open each project, agent or Settings view they land on, and a relaunch reopens the last row they opened. Enter gives an agent's pane the keyboard.
  • For --json readers:
    • fleet pause --json and fleet resume --json now print the bulk report restart --all prints (receipts, summary, ok, with signals and kept), not 0.9.1's {"paused", "project", …} object.
    • A fleet agent's state can now be paused, and its row carries paused_at.
    • project rename --json adds a notes list.
    • aisquare instances --json is new: {"instances": [...]}.
    • The upgrade --json and uninstall --json plans add asqs, and their refusal while an asq runs is "error": "asq_running".
    • Under accounts usage --json, a reading the Mac's asq took for an SSH session carries "recorded_elsewhere": true, and a usage cause may be keychain_locked_ssh, or token_refreshing while a stale token is refreshed, with refresh_retry_at once its first try has failed.
    • fleet capacity --json may give max_agents_per_project as the string "auto", and adds busy, busy_floor, busy_burst, busy_hold_minutes, signals, holds and bursts.
    • A spawn that auto admission holds fails with "error": "capacity_busy", and its hint names the retry_at.
    • metrics stats --json gives a repository's real name, not its nickname, as repos.rows[].name and repos.top.name.
    • aisquare outcomes --json and aisquare fleet top --json are new.

Known issues: none of these blocks 0.9.2.

  • Renames move no worktree or branch: a renamed agent keeps working in the tree named for its old label, and a renamed project's branches stay fleet/<old codename>/….
  • Only asq's start looks for a 0.9.1 asq, and it can miss one. aisquare upgrade, aisquare uninstall, aisquare instances and the chip see asqs of 0.9.2 or later only. asq's start misses a 0.9.1 asq started less than 10 seconds before it, one whose path has a space in it, and any on native Windows, where there is no ps to ask. Quit old asqs yourself before you update or uninstall.
  • Spawns started at the same moment can still race at the edges. Several fleet spawns at once into a project whose tmux session does not exist yet can lose one to tmux's duplicate session error, and that spawn leaves nothing behind. A number in max_agents_per_project, max_agents_total, or auto admission's floor on a busy machine can let one spawn too many through, when two spawns write their rows in the opposite order to their timestamps. The burst guard on a calm machine settles by write order and is exact.
  • aisquare outcomes can credit a rare tell to the wrong sender. The markers that record who started a turn are matched by their text across the machine, not per pane. If a paste times out, its marker can be taken by an identical prompt in another session within 30 minutes.
  • At 160 columns most real email addresses end in … in the Accounts table, with the whole address on hover. The 7-cell plan column still cuts enterprise, and it has no tooltip.

The full list, with every detail, is in CHANGELOG.md → 0.9.2.

Full diff: v0.9.1...v0.9.2