You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is an open space for the community to share feedback, ideas, and feature suggestions for Compozy.
We'd love to hear from you:
What do you like or dislike about the current experience?
What features or improvements would you like to see?
Any workflow ideas that could make Compozy more useful?
How to contribute
Simply leave a comment below with your thoughts. Be as detailed as you'd like — context helps us prioritize better.
Bug reports don't belong here
If you've found a bug, error, or unexpected behavior, please do not post it here. Instead, open a new issue with steps to reproduce the problem. Bug reports mixed into this thread will be harder to track and may get lost.
I suggest removing the step-by-step approval for the PRD and Tech Spec. Instead, generate the full preview first and ask for changes at the end. I think most people will naturally click “Next, Next, Finish” and only review everything afterward.
Reacted by Pedro Nauck, Hebert Albertin, OldBoyDev, Bruno Guimarães and rjsoux
I suggest removing the step-by-step approval for the PRD and Tech Spec. Instead, generate the full preview first and ask for changes at the end. I think most people will naturally click “Next, Next, Finish” and only review everything afterward.
Maybe make this optional. This was one of the things I noticed I enjoyed more compared to the previous Compozy
I suggest removing the step-by-step approval for the PRD and Tech Spec. Instead, generate the full preview first and ask for changes at the end. I think most people will naturally click “Next, Next, Finish” and only review everything afterward.
I kind of liked that new way of confirmation. So I agree with @camopy in making it optional being the best solution
EDIT: I noticed the .compozy folder should be in the .gitignore, right? This suggestion doesn't make sense.
What about supporting a counter code prefix for the tasks? I manually added a number as a prefix for the folders, and the IA starts working. But it would be simpler if we could refer to the tasks/reviews by the task/ticket number in the config file:
I'm working in a feature for running multiples tasks in the same command, like this: compozy tasks run-multiple <task_slug_1>,<task_slug_2>,<task_slug_NN>. The idea for the v1 is run in sequence, and for the v2 to make it run in parallel, but with git worktrees, I would like to know if its valid or not for me to continue. Here you can find the PRD and TechSpec: multi-task-run
Title: Parallel runs across projects — the one thing forcing me to step away (for now)
Hi! First, congratulations on Compozy. I'm a solo founder running several ventures, and I've used Compozy end-to-end (PRD → TechSpec → Tasks → Execution → Reviews) on a number of small projects with great success. The artifact-first, markdown-everywhere design is exactly how I believe AI-assisted development should work, and the workflow memory system is genuinely underrated. So thank you for building this.
That said, I've hit one limitation that is, unfortunately, forcing me to pause my use of Compozy — and I wanted to explain it properly rather than just disappear, because I'd love to come back.
My use case
I run multiple small, fully independent projects (separate repos, separate workspaces). To save wall-clock time, I run them in parallel — each one is its own Compozy workflow, executed by Claude Code. Individually, each run works beautifully. Together, they don't.
What I observed
When I start runs from different repos at the same time, they all hit the same daemon and serialize. I dug into why:
The daemon is a singleton per $HOME. I verified empirically that compozy daemon status from three different directories (two unrelated repos plus /tmp) reports the same socket: ~/.compozy/daemon/daemon.sock. So separate repos, tmux sessions, or git worktrees don't help — they all talk to the same engine.
Worse, a second concurrent start from a different workspace didn't queue gracefully in my case — it failed with Internal Server Error, which made me lose track of runs a few times.
Reading the source, I found that this seems intentional-for-now rather than an oversight:
internal/core/workspace/config_types.go already defines both TaskRunMultipleModeEnqueued and TaskRunMultipleModeParallel, and config validation accepts parallel as a valid value.
But internal/daemon/task_multi.go currently rejects it with parallel run_multiple mode is not supported by the daemon; use enqueued.
Meanwhile RunManager already tracks multiple active runs (ActiveRunCount, ActiveRunCountsByMode), so the skeleton for concurrency appears to be in place.
In other words: it looks like you've already designed the door — it just isn't open yet. That's actually what makes me optimistic enough to write this.
What I'm asking for (two tiers)
Tier 1 — small lift, unblocks me immediately: an official COMPOZY_HOME environment variable (or --home flag) to override the home directory resolved in internal/config/home.go. Today I work around the singleton with a "mirror home" (a fake $HOME that symlinks everything except .compozy), which works because os.UserHomeDir() reads $HOME on Unix — but that's an unsupported hack that could silently break on any release. An official override would let users run one daemon per project deliberately, with a stable contract. I'd guess this is a small change, and I'd be happy to submit a PR for it (with tests following the existing home_test.go patterns) if you're open to the approach.
Tier 2 — the real fix: implement the parallel mode the schema already promises, so a single daemon can execute runs from independent workspaces concurrently. Even a conservative version (parallelism across workspaces only, keeping per-workspace runs enqueued) would cover my entire use case and, I suspect, that of many solo builders juggling several small projects.
And independently of both: if concurrency isn't supported yet, a clear 409 Conflict / "run queued behind X" response instead of Internal Server Error would make the current behavior much easier to live with.
Where I stand
For now I have to move my parallel workloads elsewhere, which I'm genuinely not happy about — nothing else I've evaluated matches Compozy's spec pipeline. I'll keep the project starred and watched, I'm glad to beta-test any branch that touches this, and as said above, I'm willing to contribute the Tier 1 PR myself if you'd accept it.
Thanks again for the tool — looking forward to coming back.
I suggest removing the step-by-step approval for the PRD and Tech Spec. Instead, generate the full preview first and ask for changes at the end. I think most people will naturally click “Next, Next, Finish” and only review everything afterward.
Additional to this, I noticed sometimes the prd and tech-spech are not shown anywhere to the user when the final approval is asked, since the agents need aproval before saving the .md file. I was working on a similar tool when I discovered compozy and will try to create a pull request on this idea and submit it for discussion.
Seria muito útil poder anexar imagens diretamente na sessão de chat do Compozy (durante a interação com o agente), e não apenas nos comentários do GitHub. Isso ajudaria bastante em casos como:
Enviar screenshots de erros, stack traces ou comportamentos inesperados da UI para o agente analisar visualmente.
Anexar diagramas de arquitetura, wireframes ou mockups como referência ao gerar PRD/TechSpec.
Compartilhar prints de logs, dashboards ou telas de terceiros que não podem ser copiados como texto.
Dar contexto visual para tarefas de front-end (ex: "implemente esse layout" com print da imagem anexa).
Como sugestão de implementação: suporte a drag-and-drop ou paste de imagem no input do chat (similar ao GitHub), com o agente processando via modelo multimodal (Claude, GPT-4V, etc.) quando disponível. Isso se encaixaria bem no fluxo baseado em artefatos/Markdown que já existe, já que a imagem poderia ser salva como referência dentro do workspace do task.
In the Aug 10 stream I saw the Tailscale extension for connecting machines. Any chance of supporting WireGuard too? I ran into some issues with Tailscale on my Mac and prefer plain WireGuard.
In the Aug 10 stream I saw the Tailscale extension for connecting machines. Any chance of supporting WireGuard too? I ran into some issues with Tailscale on my Mac and prefer plain WireGuard.
Seria muito útil poder anexar imagens diretamente na sessão de chat do Compozy (durante a interação com o agente), e não apenas nos comentários do GitHub. Isso ajudaria bastante em casos como:
Enviar screenshots de erros, stack traces ou comportamentos inesperados da UI para o agente analisar visualmente.
Anexar diagramas de arquitetura, wireframes ou mockups como referência ao gerar PRD/TechSpec.
Compartilhar prints de logs, dashboards ou telas de terceiros que não podem ser copiados como texto.
Dar contexto visual para tarefas de front-end (ex: "implemente esse layout" com print da imagem anexa).
Como sugestão de implementação: suporte a drag-and-drop ou paste de imagem no input do chat (similar ao GitHub), com o agente processando via modelo multimodal (Claude, GPT-4V, etc.) quando disponível. Isso se encaixaria bem no fluxo baseado em artefatos/Markdown que já existe, já que a imagem poderia ser salva como referência dentro do workspace do task.
Ótimo sugestão @pedrotecinf você poderia abrir uma issue com ela?
Hi Pedro. A process question rather than a suggestion, and I would rather ask it here than guess.
Over the past days I ran a structured comparison of agent-orchestration tools with CompozyOS as one of
them, on a dedicated Linux box, and it surfaced eight findings on 0.3.0-beta.23. I filed them a few
minutes ago, routing each one by what your own configuration says. ISSUE_TEMPLATE/config.yml states the issue tracker
is for bugs and planned work, and the feature template says to start in Ideas while an idea is still taking
shape:
Discussions has no other threads, so I suspect those two are easy to miss. That is the main reason for
this comment.
And the question: was that the routing you would have wanted? If you would rather have the whole set as
a single thread, batched differently, prioritised by what actually matters to you, or closed entirely, say
so and I will do it. I have no attachment to the current shape. I am happy to re-run any measurement with
a different arrangement, and to open a pull request for any of them.
Three further findings I held back deliberately: one only bites installs that use the review trail for
approvals, one I cannot reproduce on the latest release so it fails your own checklist, and one is going to
the private security channel rather than a public issue.
Thanks for building this in the open. The --dry-run provenance output and compozy doctor were what
made several of these findings possible to state precisely instead of vaguely.
Activity
[-]Feedbacks & Suggestions[/-][+]✨ Feedbacks & Suggestions[/+]rodrigobranas commentedon Apr 3, 2026
I suggest removing the step-by-step approval for the PRD and Tech Spec. Instead, generate the full preview first and ask for changes at the end. I think most people will naturally click “Next, Next, Finish” and only review everything afterward.
camopy commentedon Apr 4, 2026
Maybe make this optional. This was one of the things I noticed I enjoyed more compared to the previous Compozy
zamorano-dev commentedon Apr 5, 2026
I kind of liked that new way of confirmation. So I agree with @camopy in making it optional being the best solution
brunomguimaraes commentedon Apr 14, 2026
I'd like to have some sort of feedback when at the end of a lifecycle. Simple sound could work.
I've drafted something here
iagowp-mthr commentedon Apr 30, 2026
The link to create a new issue is broken
antoniofulg commentedon May 12, 2026
EDIT: I noticed the .compozy folder should be in the .gitignore, right? This suggestion doesn't make sense.
What about supporting a counter code prefix for the tasks? I manually added a number as a prefix for the folders, and the IA starts working. But it would be simpler if we could refer to the tasks/reviews by the task/ticket number in the config file:
MatheusBBarni commentedon May 17, 2026
I'm working in a feature for running multiples tasks in the same command, like this:
compozy tasks run-multiple <task_slug_1>,<task_slug_2>,<task_slug_NN>. The idea for the v1 is run in sequence, and for the v2 to make it run in parallel, but with git worktrees, I would like to know if its valid or not for me to continue. Here you can find the PRD and TechSpec: multi-task-runrjsoux commentedon Jun 11, 2026
Title: Parallel runs across projects — the one thing forcing me to step away (for now)
Hi! First, congratulations on Compozy. I'm a solo founder running several ventures, and I've used Compozy end-to-end (PRD → TechSpec → Tasks → Execution → Reviews) on a number of small projects with great success. The artifact-first, markdown-everywhere design is exactly how I believe AI-assisted development should work, and the workflow memory system is genuinely underrated. So thank you for building this.
That said, I've hit one limitation that is, unfortunately, forcing me to pause my use of Compozy — and I wanted to explain it properly rather than just disappear, because I'd love to come back.
My use case
I run multiple small, fully independent projects (separate repos, separate workspaces). To save wall-clock time, I run them in parallel — each one is its own Compozy workflow, executed by Claude Code. Individually, each run works beautifully. Together, they don't.
What I observed
When I start runs from different repos at the same time, they all hit the same daemon and serialize. I dug into why:
$HOME. I verified empirically thatcompozy daemon statusfrom three different directories (two unrelated repos plus/tmp) reports the same socket:~/.compozy/daemon/daemon.sock. So separate repos, tmux sessions, or git worktrees don't help — they all talk to the same engine.Internal Server Error, which made me lose track of runs a few times.Reading the source, I found that this seems intentional-for-now rather than an oversight:
internal/core/workspace/config_types.goalready defines bothTaskRunMultipleModeEnqueuedandTaskRunMultipleModeParallel, and config validation acceptsparallelas a valid value.internal/daemon/task_multi.gocurrently rejects it withparallel run_multiple mode is not supported by the daemon; use enqueued.RunManageralready tracks multiple active runs (ActiveRunCount,ActiveRunCountsByMode), so the skeleton for concurrency appears to be in place.In other words: it looks like you've already designed the door — it just isn't open yet. That's actually what makes me optimistic enough to write this.
What I'm asking for (two tiers)
Tier 1 — small lift, unblocks me immediately: an official
COMPOZY_HOMEenvironment variable (or--homeflag) to override the home directory resolved ininternal/config/home.go. Today I work around the singleton with a "mirror home" (a fake$HOMEthat symlinks everything except.compozy), which works becauseos.UserHomeDir()reads$HOMEon Unix — but that's an unsupported hack that could silently break on any release. An official override would let users run one daemon per project deliberately, with a stable contract. I'd guess this is a small change, and I'd be happy to submit a PR for it (with tests following the existinghome_test.gopatterns) if you're open to the approach.Tier 2 — the real fix: implement the
parallelmode the schema already promises, so a single daemon can execute runs from independent workspaces concurrently. Even a conservative version (parallelism across workspaces only, keeping per-workspace runs enqueued) would cover my entire use case and, I suspect, that of many solo builders juggling several small projects.And independently of both: if concurrency isn't supported yet, a clear
409 Conflict/ "run queued behind X" response instead ofInternal Server Errorwould make the current behavior much easier to live with.Where I stand
For now I have to move my parallel workloads elsewhere, which I'm genuinely not happy about — nothing else I've evaluated matches Compozy's spec pipeline. I'll keep the project starred and watched, I'm glad to beta-test any branch that touches this, and as said above, I'm willing to contribute the Tier 1 PR myself if you'd accept it.
Thanks again for the tool — looking forward to coming back.
rafiti052 commentedon Jun 18, 2026
Additional to this, I noticed sometimes the prd and tech-spech are not shown anywhere to the user when the final approval is asked, since the agents need aproval before saving the .md file. I was working on a similar tool when I discovered compozy and will try to create a pull request on this idea and submit it for discussion.
andreustimm commentedon Jun 24, 2026
Hi Pedro, I'm from Brazil, and I have a couple of questions.
Do we have the feature to work with Git worktrees in Compozy today?
Is there a timeline for releasing this feature?
Do you need help building it?
pedronauck commentedon Jun 26, 2026
Hi Andre, yes for sure, it's about to be released a new worktree feature indeed.
feat: support COMPOZY_HOME env override for home directory (#216)
pedrotecinf commentedon Aug 12, 2026
Suporte a anexar imagens na sessão de chat
Seria muito útil poder anexar imagens diretamente na sessão de chat do Compozy (durante a interação com o agente), e não apenas nos comentários do GitHub. Isso ajudaria bastante em casos como:
Como sugestão de implementação: suporte a drag-and-drop ou paste de imagem no input do chat (similar ao GitHub), com o agente processando via modelo multimodal (Claude, GPT-4V, etc.) quando disponível. Isso se encaixaria bem no fluxo baseado em artefatos/Markdown que já existe, já que a imagem poderia ser salva como referência dentro do workspace do task.
rjsoux commentedon Aug 12, 2026
Hi Pedro, congrats on Compozy!
In the Aug 10 stream I saw the Tailscale extension for connecting machines. Any chance of supporting WireGuard too? I ran into some issues with Tailscale on my Mac and prefer plain WireGuard.
Thanks!
pedronauck commentedon Aug 12, 2026
Of course, can you please open an issue 🙏🏻
pedronauck commentedon Aug 12, 2026
Ótimo sugestão @pedrotecinf você poderia abrir uma issue com ela?
GabrielKifer commentedon Aug 14, 2026
Aqui do lado Windows da força, só esperando a vez pra usar essa maravilha 😅 @pedronauck
jillesmc commentedon Sep 9, 2026
Hi Pedro. A process question rather than a suggestion, and I would rather ask it here than guess.
Over the past days I ran a structured comparison of agent-orchestration tools with CompozyOS as one of
them, on a dedicated Linux box, and it surfaced eight findings on
0.3.0-beta.23. I filed them a fewminutes ago, routing each one by what your own configuration says.
ISSUE_TEMPLATE/config.ymlstates the issue trackeris for bugs and planned work, and the feature template says to start in Ideas while an idea is still taking
shape:
file:line,verified byte-identical on
main, and a reproduction that costs you nothing to runI can show a busy loop that burned 25 h of CPU, but I cannot reproduce it on demand) and Is the 30-minute silent quiet_after window meant to replace the 15-minute runtime_warning? #568 in Q&A
(whether the 30-minute silent
quiet_afterwindow was meant to replace the 15-minuteruntime_warning; that one turns on your design intent, not on a defect I can demonstrate)Discussions has no other threads, so I suspect those two are easy to miss. That is the main reason for
this comment.
And the question: was that the routing you would have wanted? If you would rather have the whole set as
a single thread, batched differently, prioritised by what actually matters to you, or closed entirely, say
so and I will do it. I have no attachment to the current shape. I am happy to re-run any measurement with
a different arrangement, and to open a pull request for any of them.
Three further findings I held back deliberately: one only bites installs that use the review trail for
approvals, one I cannot reproduce on the latest release so it fails your own checklist, and one is going to
the private security channel rather than a public issue.
Thanks for building this in the open. The
--dry-runprovenance output andcompozy doctorwere whatmade several of these findings possible to state precisely instead of vaguely.