Skip to content

Releases

Foreman 1.2.5

Choose a tag to compare

@github-actions github-actions released this 18 Nov 06:15
31b859a

Changed

  • Opening a channel takes you to its newest post and the message box instead of its first post. (#222)
  • Every AI employee gets a shell by default (turn it off in Edit) and every MCP server a workspace's agent has, and can push and open pull requests itself, with nothing waiting for an approval when unattended. (#223)
  • AI employees have no limits: no runs a day, minutes, cost threshold, one-run-at-a-time rule, pause after failed runs, hand-off limits or minimum time between scheduled runs. A chain of hand-offs still stops after 20 runs until you raise it. (#223)
  • The sidebar is decluttered: Find is a ⌕ in the top row that opens the search box in place, and everything you can add (workspace, repository, employee, channel, automation, organization) is in one + menu. Every section header folds the same way, and Team page and All channels are links under their lists. (#224)
  • In the sidebar, the team shows as a row of hats, and channels and workspaces take one line each; hover or Tab to a workspace to see its repository, branch and age. (#224)

Fixed

  • A folded comparison or flow-run group in the sidebar shows ▲ when something inside needs you. (#224)

Known validation findings

npm audit reports a high-severity advisory in picomatch, a regular-expression denial of service through repeated extglobs (GHSA-fc5q-39xc-vjpf). It covers every picomatch version, so no fixed version exists yet; Foreman's own code reaches it only through the experimental Daytona sandbox provider's SDK, and otherwise through build and test tooling.

One test failed on the Windows machine that checked this release, as it did for 1.2.4: tests/employees.test.ts › "gives every employee its own prompt file…" compares Foreman's prompt, written with plain line ends, against a test fixture that a Windows checkout with core.autocrlf gives CRLF line ends. The behaviour it checks is right. Four other test failures in the full run passed when rerun with fewer workers.

Full changelog: v1.2.4...v1.2.5

Foreman 1.2.4

Choose a tag to compare

@github-actions github-actions released this 17 Nov 08:52
a9763a0

Added

  • A project part that designs screens draws them as HTML mocks, two options each, in the project's mocks folder, and the channel shows a button for each that opens it in your browser; the UX designer (Uma) now draws mocks for what she proposes. (#219)
  • A mock an AI employee saves in memory shows in its channel or task as a button, such as Open sidebar-simplify/index.html in your browser, that opens it in your browser, and the Memory view has Open in browser for an HTML file. (#220)

Changed

  • Every AI employee, including Priya and the job templates, can change its organization's repositories, push and open pull requests without waiting for you. Existing employees are updated once, and keep their approval. (#211)
  • Runs a day and Minutes a run are optional and empty by default: an employee runs as often and as long as it needs unless you set a limit. (#211)
  • Every job template employee, including Priya, follows a prompt written for its job and gets better whenever a newer Foreman improves that prompt, without you approving it again. An employee's prompt is now its own file (bots/<id>.md); for a template employee, Edit shows the job's prompt and lets you add only what your organization wants on top. (#215)

Fixed

  • On a project's questions card, Send answers no longer stays greyed out when an answer is missing: it goes to the question you missed and names any others. (#213)
  • Clicking a Foreman notification on Windows brings Foreman to the front again, restoring it when minimized. While Foreman is closed and agents keep running, clicking one opens Foreman. (#214)
  • A project's workspace no longer reappears in the sidebar's Needs your attention after each run, and what a project waits on is shown only in its channel. (#216)
  • A note written at a project's check and sent with Approve with note now reaches the parts that start next, and a note on the plan or final check can only be sent with Send back, so it's never silently dropped. (#218)
  • Foreman deletes temporary files it left behind after a crash, once they are a day old. (#217)

Development

  • Test runs no longer leave folders in the system temp folder. (#217)

Known validation findings

One test failed on the Windows machine that checked this release: tests/employees.test.ts › "gives every employee its own prompt file…" compares Foreman's prompt, written with plain line ends, against a test fixture that a Windows checkout with core.autocrlf gives CRLF line ends. The behaviour it checks is right. Nine other test failures in the full run passed when rerun with fewer workers.

Full changelog: v1.2.3...v1.2.4

Foreman 1.2.3

Choose a tag to compare

@github-actions github-actions released this 16 Nov 22:16
1c6f159

Added

  • New employee opens on Pick a job: choose one of Foreman's fifteen jobs and Add makes that employee approved and ready at once; Write your own keeps the Generate way. (#204)
  • The program manager asks you her questions before she plans, all on one card: answer each, or tell her to Plan with what you have. For a large project she assumes nothing she could ask. (#205)
  • A project's plan is drawn as a workflow, with what the program manager understood, when it is done, what is out of scope and any assumptions to check, and plan.md carries the same, with the questions and your answers. (#205)
  • A project's Parts card can show the parts as a map (List | Map): what runs at the same time, what waits for what, and where each part is; selecting a part takes the channel to its newest post. (#208)
  • The Team page shows Hand-offs outside projects: each chain of work one employee handed to another, drawn as a map with the actions waiting for you on it. (#208)
  • A project's channel shows who is at work on it now, on what and for how long, with Watch to open their chat, or Nothing running when no one is. Its row in the sidebar and on the Channels page shows a turning ring while someone works. (#209)
  • The sidebar's groups, such as Needs your attention and Agents working, now fold under cards like Channels, remembered across launches; a folded Needs your attention still shows how many need you. (#203)

Changed

  • Employees picked from a job template follow a prompt written for their job: how it is done, what they hand back, when they ask, and they no longer claim tests passed when nothing ran. (#210)
  • A project started in a channel keeps its workspace in the channel. It no longer appears under the sidebar's or Home's workspaces unless it needs you. (#207)
  • Once a plan is in effect, its table shows each part's State. (#208)

Fixed

  • An organization's project no longer stops with "Maximum of 10 chats per workspace" after about ten employee runs; a workspace can now keep up to 100 chats open. (#206)

Development

  • A project-parts screen in the screenshot fixture. (#208)

Known validation findings

npm run check:security did not finish its scan of Git history: Gitleaks stopped at its 5-minute limit after scanning about 4.3 GB across all refs, with no leaks found in what it scanned. The source and lockfile scans found no secrets.

Full changelog: v1.2.2...v1.2.3

Foreman 1.2.2

Choose a tag to compare

@github-actions github-actions released this 16 Nov 10:41
c473cb2

Added

  • New organization lets you pick ready employees as well as Priya, from fifteen job templates such as software engineer and UX designer: press a job's chip to add them, hover or focus it to see what the job does, or open Compare jobs… to see them all as cards. They're approved with the organization. (#201)

Full changelog: v1.2.1...v1.2.2

Foreman 1.2.1

Choose a tag to compare

@github-actions github-actions released this 16 Nov 08:06
0326139

Added

  • The Channels row has a + that starts a new channel with @Priya, and each channel has a menu to Rename, Archive or Delete it; deleting one still going stops it first, and its files stay unless you choose otherwise. (#199)
  • All channels opens a page with every channel of the organization, archived ones too, to search, filter, and archive or delete several at once. (#199)
  • New organization can Keep its repositories in one folder: its memory repository is made there, and while the organization is showing, Clone and Quick start put new repositories there and add them to it. (#197)

Changed

  • A closed channel is archived after two weeks instead of disappearing, and can be restored. (#199)
  • An agent's question card is about half as tall: the question is its title, each tab shows your answer, and the options are a compact list with their descriptions beside them. (#196)

Full changelog: v1.2.0...v1.2.1

Foreman 1.2.0

Choose a tag to compare

@github-actions github-actions released this 16 Nov 03:14
3d91ea1

Added

  • Every organization has a program manager, Priya. Tag @Priya in the organization's channel under Channels to start a project: she plans it, proposes employees when nobody fits, and the parts run with your checks where she plans them. (#194)
  • A project's channel shows its plan, each part's result and file, and what waits for you; @Name there asks anyone who worked on it a follow-up. (#194)
  • Hand to Priya… in a workspace's header hands her a workspace that grew into a project, with its chat and branch. (#194)
  • Ask a follow-up on an employee's finished task. (#194)

Changed

  • New organization shows Your team: you and Priya. (#194)

Full changelog: v1.1.1...v1.2.0

Foreman 1.1.1

Choose a tag to compare

@github-actions github-actions released this 15 Nov 11:00
826287a

Changed

  • While an agent works, the message box's button stays Steer or Queue and Stop is a small ■ beside it, appearing a moment into the turn, so a double-click on send can no longer stop the turn it just started. Esc still stops at once. (#190)
  • The list of files and commands under a turn's Completed line starts folded on every turn, the latest one too; click the line to open it. (#191)

Full changelog: v1.1.0...v1.1.1

Foreman 1.1.0

Choose a tag to compare

@github-actions github-actions released this 15 Nov 09:03
fea1e5c

Added

  • AI employees: organizations with their own memory repository, and employees you describe in a sentence, approve, and ask for work, or start on a schedule or an event. They work in fresh workspaces, remember what they learn, hand work to each other, and hold pushes, pull requests, Slack posts and Linear changes for you in Needs you. Make an organization from the Personal ▾ switcher at the top of the sidebar: New organization… asks only for a name and which repositories to bring. (#183, #184)

Changed

  • When an agent asks for an approval, a question or a plan review, its card docks above the message box, so what you were typing stays, and Enter in the box never answers it; F6 moves to the card. (#187)
  • An approval spells out a command part by part, marks anything that leaves this computer, and says exactly what Always allow adds. (#187)
  • Background shells and subagents now show in one block above the composer: what each one is doing, how many are done, the time the turn has run, a subagent's Latest steps, and Stop on a subagent. (#186)
  • One count of what needs you per space, shared by the sidebar, the rail, the space switcher, Team and Home, and held pushes and memory proposals are included. (#187)
  • The sidebar's space switcher is its heading, Automations and Assistant share a row, Team is one row, and employee runs show their hard hat; the list starts much higher. (#187)
  • Home leads with what needs you, filtered by space, and folds Get started to one row. (#187)
  • Each turn ends with a receipt of its outcome, files and commands; a failed turn leads with its cause and Retry. (#187)
  • The send button says what Enter will do (Send, Steer, Queue), and the context meter shows what's left before compacting. (#187)
  • The Team page shows a column per employee; an employee's page and its run show its decisions; a draft employee is approved from its own page. (#187)
  • Settings shows one short page per menu entry, with Previous / Next, search results with current values, and a Devices category. (#187)
  • A flow run's panel opens with the flow's name, how it ended, its time and cost, and the graph's loops and zoom in a bar above it. (#187)
  • In a narrow window or at 200% zoom the workspace list folds to the rail and the side panel opens as a sheet. (#187)

Fixed

  • New workspace and Ctrl+N no longer open on an organization's memory repository or another space's repository. (#187)
  • Plan review's Ctrl+Shift+Enter approves as its label says; Esc with an empty @ or / menu no longer stops the agent. (#187)
  • Windows shows shortcut keys as words (Ctrl+Shift+Enter) instead of Mac symbols. (#187)
  • The model picker's rows no longer run together, a count of one reads "1 needs you", a search with no matches no longer says "1 match", diffs no longer break words mid-identifier, Run a flow no longer lists a model twice, and the sidebar no longer scrolls sideways. (#187)
  • "Waiting for N background tasks to finish." now shows in the app's language. (#186)
  • On Windows, a workspace archived after an AI employee's run, a scheduled run or a merge no longer stays behind half-deleted when another program briefly holds its folder. (#188)
  • On Windows, a settings change made while another was being saved is no longer occasionally lost. (#188)

Development

  • The design system gains HardHat, the avatar AI employees wear: the Foreman hard hat in one of twelve colours. (#182)
  • The screenshot fixture and npm run dev demo include an organization with AI employees and their work, and seven new fixture screens show them. (#185)
  • In npm run dev demo, a chat message that mentions "background" plays a turn with a background shell and subagent. (#186)
  • A ui-review skill (.claude/skills/ui-review) audits the app's screens against a rubric and mocks improvements. (#187)

Full changelog: v1.0.23...v1.1.0

Foreman 1.0.23

Choose a tag to compare

@github-actions github-actions released this 13 Nov 14:31
5a9d1a7

What's Changed

  • Show a background agent resumed with SendMessage as a subagent row, with Stop and what it did since by @YatinChaubal in #178
  • Scroll to Limits and Usage when they're picked in Settings, instead of staying on Data by @YatinChaubal in #179
  • Keep a timed-out switch-agent test from failing the next one, and give its turn-heavy tests room under load by @YatinChaubal in #181
  • Release 1.0.23 by @YatinChaubal in #180

Full Changelog: v1.0.22...v1.0.23

Foreman 1.0.22

Choose a tag to compare

@github-actions github-actions released this 12 Nov 09:17
ae3e859

Added

  • Accounts in Settings → Agents for Claude Code, Codex and Cursor: more than one account per agent (such as work and personal), each with its own sign-in, config folder and key, chosen in the model picker's Account row, with a rule for whether a chat may move between accounts. (#161)
  • Settings → Agents shows each agent's capabilities: what works, what works another way, what isn't offered, and who enforces its permissions. (#173)
  • Usage → Limits shows each account's usage windows. A chat that hits a limit waits and resumes by itself when the limit resets, and can Continue on another account. (#173)
  • Switching agent or account in a chat shows What Codex received (or another agent's name), and agents can read the turns that were left out. (#173)
  • Subagents appear under their turn with their own Stop, and their approvals say which subagent asked. (#173)
  • Codex can plan: turn on Plan, review its plan, and it carries on to implement it once you approve. (#162)
  • Messages you send while Codex works now reach it in the running turn, instead of waiting for it to finish. (#162)
  • Sign in with ChatGPT in Settings › Agents › Codex signs in through the browser; the terminal stays as Use a terminal instead. (#162)
  • Codex chats can be forked, rewound, compacted, and given a goal. (#173)
  • Claude Code and Codex sessions on this computer can sign in to Foreman over MCP and start and message work, wait for turns and answer their workers' questions, at an access level you choose when you approve them in Settings → Data & devices → Agents on this computer. (#170)
  • An outside agent at Auto or Never ask can propose a schedule, which waits in Automations until you approve it. (#170)
  • foreman service install starts Foreman with no window whenever you log in, so schedules, flows and phones keep working without opening the app; foreman service and foreman service uninstall show and remove it, and foreman service restart stops it and starts it again, then waits for it to answer (--force restarts even with agents running). (#167, #169)
  • Foreman started by the background service installs a downloaded update by itself once nothing is running and no window is open, then starts again in the background. It copies your data aside first: if the new version can't open it and start, your data is put back as it was, foreman service says how the last update went, and a version that failed isn't installed by itself again. (#167, #169)
  • Run Foreman without a window with --headless, or keep it running in the tray after the last window closes with Keep running without a window. (#163)
  • Open the whole Foreman in a browser at /app on the phone's address, signed in from the phone dialog's Browser choice or foreman phone link --browser. (#163)
  • Pairing a phone now chooses what it may do: Watch, Operate or Full control. (#163)
  • foreman phone signs phones and browsers in from a terminal: status, link, pending, approve, deny, revoke. (#163)
  • The phone updates live as agents work instead of refreshing every few seconds, shows a reply as it is written, and says when it is reconnecting or has been signed out. (#163)
  • A paired phone can Steer a running turn or Queue a message, and Send now or Delete queued messages. (#166)
  • A paired phone can comment on a line in Changes, a removed line too (the comment quotes it for the agent), delete its comments and Resolve the agent's. The comments go with the next message. (#166, #168)
  • A paired phone has a read-only Terminal tab that follows the workspace's terminals and run scripts as they print. (#166)
  • The phone's workspace menu shows how full the chat's context is, and the turns, agent time and cost of the chat and the workspace. (#166)
  • After a usage limit, a missed resume on the phone offers Resume now and Dismiss. (#166)

Changed

  • With Codex tool approvals set to Sandboxed, Codex asks before leaving its sandbox instead of being stopped there. (#162)
  • Cursor, OpenCode and chat agents can Interrupt and send a follow-up, and their questions and terminal commands appear in the chat. (#173)
  • An agent that fails to start says why in one sentence, and Settings shows whether its version was tested, with Update. (#173)
  • A phone's Terminal tab shows each terminal as its tab on the computer does, so progress bars, Windows terminal repaints and full-screen programs such as less read correctly. (#168)

Fixed

  • Rewinding a chat on Cursor, OpenCode or a chat agent keeps the conversation up to that point instead of starting with no history. (#173)
  • While set to Steer, a phone no longer says your diff comments go with the message; they wait for the next turn. (#168)

Security

  • Pairing links and codes are shown only to devices with full control. (#163)

Development

  • Adapter tests replay recorded agent traffic (npm run record:agent), and model-facing text has hand-run evals (npm run evals). (#173)
  • The Windows release build can be signed through SignPath Foundation once Foreman is approved; until then the Windows installer stays unsigned, and CODE_SIGNING.md explains the policy and how to verify a download. (#175)
  • npm run check:security passes again: it no longer flags the recorded agent fixtures' message IDs or the sanitizer test's dummy JWT. The persona template test passes on Windows checkouts that use CRLF line endings. (#176)

Full changelog: v1.0.21...v1.0.22