Also shipping this release:
- Workspaces from remote branches and PRs
- Full file browser CRUD with batch ops
- JetBrains IDE support
- Native notification deep links to agent terminals
- Tab drag-and-drop reordering
- Tons more
We're a few weeks away from public launch. A LOT more coming.
CMD+SHIFT+N to open Task Capture and create a new task for the Inbox.
✅ Tasks Board
Rebuilt with flexible grouping, inline editing, and selection-to-task from agent output.
Tasks won't replace your Jira/Linear (yet) but they are super helpful for keeping track of your individual work, and the Inbox is super helpful to make sure you don't lose track.
Our users love Subspace because it organizes their projects. Timeline is another way to help you make sense of the ebb/flow of your projects. See your conversations in a timeline view alongside file changes, commits, and more.
Also -- easily restore your comments!
⌛️ Timeline
A unified feed of everything happening in your workspace. File changes, commits, agent conversations, PR activity. Review an hour of agent work in 30 seconds.
Ship loop has massively accelerated our velocity. And because you can add custom actions to any git state, you can make the Ship Loop work exactly the way you or your team works.
www.subspace.build
🎉 New Features in v0.5.12
🚀 Ship Loop
A git state machine that watches your repo and surfaces one CTA — the right one. Commit, push, rebase, merge, open PR. When you've got 5 agents writing code across branches, you don't want to think about git. Now you don't have to.
In Subspace, you can see all your Claude Code/Codex chats, terminals, files, docs, and more in a single place. No more 12 window chaos.
Some big updates this release -- Ship Loop, Timeline, an updated Tasks Board, and tons of ux polish & bug fixes.
🔗 Link and more below!
Big @subspacehq update!
We're on a mission to make Subspace the best place to work together with Agents. Think of it as Slack for your AI team: every agent has presence, persistence, and a place to work WITH you.
makes me think true personal agents are pretty far out for most people unless one of them just builds their own versions of everything they’d normally need an api key for.
for personal agents to work for everyday people, they’ll need a way to sign up for services and get api keys for users with minimal setup. that doesn’t really exist today.
We should all start reporting the magnitude of our projects in tokens not lines of code.
The Claw is the Law
Coming soon to @subspacehq
* run your @openclaw agent(s) from any workspace or project
* view and resume OpenClaw sessions you started from that workspace
* create new OpenClaw sessions and run them side-by-side on different tasks
Fixed now in v3.5.0, including adding models to subagents to improve speed/token efficiency in high certainty scenarios.
github.com/Codename-In...
Embarrassingly in my haste to package SPECTRE for release I inadvertently .gitignored ALL scripts in the SPECTRE repo instead of the one i was targeting.
So none of the hooks worked. 🤦🏻♂️
You live, you learn.
Full scope for clarity. Smallest version for velocity.
Here's why this matters: AI is dramatically faster when it's operating on working code.
Trying to wire too many advanced features together from nothing produces chaos. A stable v0 gives you a foundation where features start compounding instead of colliding.
Find and build the smallest working version.
This matters because describing the complete feature exposes flawed assumptions you'd miss if you started small. And you can always scope down, but building without knowing where something is headed creates compounding debt.
Once the vision is clear, flip it!
Don't edit yourself yet. Brain-dump, then use the AI to help you refine and structure your thinking until it's clear.
You are not trying to get approval from a busy human engineering team. We're in a different paradigm now - engineering resources are not the bottleneck anymore.
Here's the process I keep coming back to
Start by describing exactly how you want the feature to work - the full, ideal version.
When building products with AI, you have to un-learn some things that were ingrained in you as a product builder.
Here's a really big one - *constraining scope too early due to perceived complexity*
We're building something special at Codename. I can feel it. Subspace and New June are just the beginning.
In an AI-native world, the leverage shifts from writing code to designing & maintaining systems.
SPECTRE is the structured workflow I use daily to build with AI.
Scope → Plan → Execute → Clean → Test → Rebase → Evaluate.
It’s live on Product Hunt today.
You can read all about how I actually use this day to day in the README in the repo.
You can expect them to continue to improve and I'll be adding new features as I build them.
I genuinely hope this helps people build their dreams.
It has helped me build mine.
/Evaluate - architecture review + generate product/tech docs as Skills that get auto-loaded when relevant
/Clean - lint, find dead code, find AI slop, cleanup logs, logically grouped commits
/Test - risk-aware test coverage focused on behavior not details
/Rebase (or release) - safely layer changes on a parent branch w/backup and get smoketest steps
SPECTRE is also a handy mnemonic device to easily remember the steps in the actual workflow
/Scope - get clear on what is in/out
/Plan - technical design/tasks that adapts to the size
/Execute - TDD + parallel subagents + code review + validation... one command
4. Using Skills for product/technical documentation as a form of recursive learning, allowing the Agent to auto-load relevant 'knowledge skills', reduce token usage searching the codebase, and serving as "work memory"
