The more I use Claude Projects, the more convinced I am that this thread based workflow is the right direction One of the hardest parts of working with Claude Code is managing tasks across different sessions Worktrees help, naming sessions and using colors help. But it’s still hard to... keep track of everything and visualize the history A session doesn’t just end and disappear It contains decisions, executed commands, PRs, and most importantly, a STATE: - Working - Waiting for you - Idle - Resolved Everything stays organized inside the same project, together with the Library, Pull Requests, and Routines It finally gives you one place to see what each session is doing, what happened before, and what still needs your attention My only feedback so far is that threads run inside a remote sandbox, and sometimes I need to execute commands directly on my own computer Everything else... pretty much flawless 👏show more

🫶🫶 let us know if you have feedback!! local support for threads is coming superrrr soon

@dani_avila7 I feel you on managing tasks across sessions. Hit this wall too while building my own thing. Naming sessions helps a ton.

Meh... "NADA DEL OTRO MUNDO"... Ese comportamiento ya lo he logrado replicar, hace mucho -de manera local- en la personalización auto-evolutiva que ya tengo sobre CC... Entiendo que lo hace accesible "para todos", pero PERDÉS AUTONOMÍA, porque NO es local, y además, por como está hoy CC, NO es algo difícil de lograr...

🙏 keep us posted on your thoughts! I’ve found myself mostly living inside of paired artifacts now with projects

with a real agent fleet, the transcript isn't the handoff. intent, accepted decisions, and completion evidence have to survive separately from the chat or the next session just inherits archaeology.

Un fichero de estado por rama que el agente lee al abrir cada sesión hace que el hilo sobreviva aunque cambies de máquina.

i keep learning this the expensive way

colors and names help. what's held up in my repo is a list of rules that each cost something once. dozens of files sit modified on purpose here, and this AI has learned which mess is load-bearing. the loudest rule: never git stash.

imo sessions holding all the decisions/state is the bigger problem the project should know that stuff, not the chat

Projects seem to give the work a home, while sessions are still mostly where it happened. That gap gets expensive once a task spans more than one sitting.

Treating a session as a durable state machine is the right abstraction. The missing piece is making those state transitions queryable, so a team can see why a session is waiting instead of just seeing that it stopped.

Claude code is for advance builder it takes technical quiet a bit to truly run and give us the right results

Treating each coding session as a stateful thread feels much closer to how real projects evolve. A visible history of decisions and handoffs could remove a lot of the context recovery that slows multi-session work.

Sounds like a solid structure, but the remote sandbox still feels like a chokepoint for quick local tweaks

the sandbox is the actual split. does a Waiting session still have the dirty worktree, or do you copy files down before the local command you couldn't run in the remote box?

Yep, running a few apps at once and the sessions pile up fast. Naming them helps for about a day then it's chaos again

thread per project really is the unlock. the context re-explaining tax was quietly eating an hour a day

I really am waiting for mine account to have projects, hope soon it will be available.

It hasn’t rolled out to me yet, but my current session can manage other sessions when I explicitly ask it to create a new session or archive a session.

A simple habit makes this useful for non-technical teams too: end every session by asking for three bullets—current state, next action, and blocker—and save that as the Project brief. You start the next chat with a handoff, not a history hunt.

Claude Projects solves the context window problem. The hard part isn't managing tasks, it's paying for tokens to remember what you already told it yesterday. stop renting memory start building systems

the sandbox part is the one i'd push on. i built domlabs-bot for that, telegram straight into Claude Code on my own machine, so whatever it runs hits the real repo instead of a copy.

I already have a system. I have configurable & efficient dynamic workflows (feature, review, research, feature-gathering, etc) I use a single session and the agent just launches the specific workflow for what I need. One session to handle countless tasks (threads).

keep commands local while the project thread stores state, history, and approvals.

disappointing to hear about the remote sandbox.

I still lose the plot when one Claude Code job spans three worktrees. Named sessions help. I want one project thread that owns the whole queue.

Worktrees and colors helped me right up until a fresh one confidently rebuilt the thing I killed the day before.

AI coding needs project management now 😂

managing tasks in different sessions can really slow down productivity over time

The worktree is alive after the session ends. The next tool that loads has no idea what was decided.