Hi,

This release, in numbers

More than half of these pull requests fix something that was already shipped, and there's more test than code in them.

flowforge update repairs

flowforge update used to return the moment your version matched the latest — before it re-laid the agents, hooks, commands, skills, or the ~/.claude mirror, so the only real repair door was reinstalling from scratch. It now always finishes those steps, removes the old binary before writing the new one so a bad write can't get silently killed by the code-signature check, and proves the new build runs before calling itself done. flowforge update --repair re-lays everything without the network; --force reinstalls even at parity. And flowforge doctor has a new install-freshness row that catches an install that's fallen out of sync with what's actually running.

The cockpit stays solid

On my own Mac, a cockpit with zero workers running still carried 522 dead processes and 84 dead tmux sockets — FlowForge now cleans up its own garbage at every start and prints what it removed, by name; flowforge hygiene is the manual door if you'd rather run it yourself. Your own keystrokes no longer paint themselves onto the ACE rail's header, typing no longer slows down the longer a session stays open, and an idle cockpit stops burning CPU on nothing. A paste that lands mid-composition no longer freezes the terminal, and FlowForge now asks your installed Claude Code which permission modes it accepts instead of assuming — a fix that came from a customer report.

Reading the cockpit at a glance

The header's clocks are now named for what they actually measure: FLOWFORGE SESSION survives an ACE restart, and CONTEXT zeroes on /clear and on a reset, so you can see how old the context you're trusting really is. The ctx gauge stops jumping backwards mid-session — it now shows the highest value the session has reached, not whatever line happened to replay last. And during a long turn, the ACE rail narrates what's actually happening: the current step, a ledger of what landed, and a clock per step.

Plans and the Gantt

A sprint can now depend on another sprint: flowforge sprint-dep records the edge and names the date consequence without moving anything, and the Gantt draws a whole initiative's sprints on one chart with the cross-sprint critical path. The Gantt also stops saying "on track" in green beside a ticket that's already open past its planned end — it shows the overdue count instead. And the earned-schedule reading is back on the Gantt strip.

Reviews that no longer strand a worker

When the session that dispatched a code review ended, the review crew died without writing anything — and the round said "still running" for up to three hours and ten minutes, a dead end only a person could clear. The round now checks whether the crew is still alive, records a dead one as ended with no verdict, and re-dispatches the review automatically. And the override that lets a worker push past a just-refused gate now has to name a class — base-sync: <why> or founder: <why> — instead of any sentence a worker cared to write.

Colado

Colado is a jiu-jitsu term for a position with no gaps — justo, perfectly adjusted, full pressure. It's the name we gave the FlowForge gates that leave no space between the rule and your work. This release closes five more: a worker can no longer be built on a machine that's plainly signed out of Claude Code, an update can no longer count a binary that won't start as installed, a worker can no longer grant itself the override a guard just refused, a child process FlowForge starts can no longer be left unreaped, and a dead review crew can no longer hold a round open for hours.

Small things that change everything

Not every change earns a headline. These fit in a line each, and you feel them every day.

The ForgePlay picker scrolls now, so the selection can never land on a play you can't see. F5 repaints the whole cockpit, whatever state it's stuck in. A worker that's finished its job reads retired — not vanished, which is how a crash reads. And an auto-merge names its PR, head and gate right on the rail.

One setup step

Sign in to Claude Code once if the cockpit asks: run flowforge claude login (or claude auth login), or click Sign in on the desktop's Claude Code row, then press r in the cockpit to re-check. On this upgrade, expect one last four-minute wait for your seat — the daemon you are leaving predates the fix; from the next upgrade on, the seat is freed on the way out. And if you set FF_GOLDEN_PUSH_OVERRIDE, give it a class: base-sync: <why> or founder: <why> — a plain sentence is now refused. As always, quit and relaunch every open cockpit after you update.

Not yet

The idle-CPU fix removes the clock that was spinning and cuts how often the process scan runs, but an "after" number from a live cockpit on this release hasn't been measured yet. A worker with no worktree and no Fleet row still isn't swept — killing one on a live cockpit, on a signal known to be unreliable, is a risk we're not taking yet. And flowforge claude login depends on Claude Code's current sign-in screen; if that screen changes, it falls back to the terminal path.

How this letter reaches you

This letter comes out of FlowForge's own release pack — the same tooling that keeps the release notes, the release page and this letter derived from one committed source. The same pipeline — render, export to .eml, send at promote — is in your own Settings, opt-in and off by default.

Unsubscribe from release letters · Manage preferences · Privacy

FlowForge · v1.2.25 · sheet 1/1