Ask HN: What do you all use to share context with team mates?

1 points by Satoshri ↗ HN
Problem: My cofounder and I work exclusively in Claude code. However, I've no idea about the conversation he is having with his agents. The only point of contact is what gets merged in Github, which is the contract. But there are times when it would be very useful for me to understand the decisions/conversation that led him to arrive at this juncture. All this context is lost.

The current best methods I've found are 1. Team/Org/Dept level shared-context.md in a cloud folder - who updates?, when?, how often?, what level of summarization?, are all unknown 2. External Tools to store context - pg, supabase, any vectordb, but then you have to maintain it, and all the same questions apply 3. Use git to share context

I've not found a single satisfactory way to achieve this. We both have multiple Claude code sessions open right now and we have no good way to communicate what each of us is trying to achieve in the session.

8 comments

[ 0.23 ms ] story [ 13.0 ms ] thread
Disclaimer: this is an ad :-D Me and my now co-founder ran into the exact same problem around February and ended up building leverage.computer been running our company on it across eng and other functions. Would love to show you if you’re down. Just email me eimen@leverage.computer
we had the same issue. What we do is instruct the agents to write any decisions or new knowledge to md files. The files are structured in a file system, separated by topics and the persona/agent making them. We use git to make the sync, but if you need it online - you can use confluence or a similar service. usually, it breaks for us when you have knowledge that is outside the sessions data, in the ticketing system and others. where do you usually lose the data, in the agents sessions or in the conversation outside it?
same problem as yours, any md handoff based system is simple to implement but lack integration with other tools like jira, slack, etc. and also it is another system we need to maintain now.
I've had a similar problem. I think the ideal shape of the solution depends on how you guys work and how in sync you want your AI context to be. Do you have shared rules that need to be followed? Markdown file in the git repo. Too many different places where information is stored? Only use tools that connect to the AI via MCP.

I built something to solve the "understand-the-conversation" problem for myself (chironmemory.com). I didn't want to worry about maintaining any sort of complex system where existing entries need to be updated, so it's basically an append-only self-organizing pile of info that your agents can throw whatever they want into and the recall does a good job of bringing it back later. I've got my whole git history and all my emails in there.