That implementation afaik doesn’t have anything to do with the actual trace format it’s only related to how they are tied to commits
That implementation afaik doesn’t have anything to do with the actual trace format it’s only related to how they are tied to commits
As always, if you have any issues, let us know - hop into the
Traces Discord, message or DM me or @traces.com, whatever works for you.
Get started at traces.com.
This is entirely opt-in - you can choose to install these skills at the scope of your project, for a specific agent, or globally.
Just run `traces setup` once you've installed Traces and logged in.
If it's not working, just run `traces remove skills`
Our team uses these skills to attach traces to every PR created by agents.
Now every PR has more context of what agent & model was used, how we prompted it, and how it arrived at a solution. This adds improves our conversation during review.
PR's look like this now:
Introducing Share from Anywhere for Traces
A set of skills for sharing to directly from any supported agent to traces.com. Available now for all 9 supported agents.
This small add-on has improved how we work with agents as a team. Here's how it works:
thx for taking a look and for taking the time to comment btw 🙏
starting with borrowing GitHub’s privacy model for Oauth
but you’re correct there’s probably more to do, what would you like to see here ?
Something like threads but works for pretty much any agent, we support cursor Claude opencode codex and others
I’ll have to dig into agent traces further to confirm the second q, I don’t believe the two have overlapping data tbf
that said cursor implements agent traces and we support cursor
and yes we will be providing api's and hooks to do more things with your traces, a lot more along these lines coming soon
it's not exactly the same, agent-trace is more about tracking ai contributions vs. human contributions. traces.com is a product for storing and sharing traces for open source & teams, gh for your agent conversations if you will
really appreciate the question & happy to dig in further 🙏🏽
Your feedback is super important as we're building out the Traces community.
You can follow @traces.com for updates, join our Discord: discord.gg/hR8Syd9ZcJ, or DM me personally.
I can't wait to hear from you and see what you share.
On Privacy
You can share Traces publicly, privately, or via direct link. Secrets from agent conversations are stripped at source.
Coming soon are robust privacy + org models, an open source CLI, self hosting & more.
Trust is a long running conversation we are just starting.
Why would I share a Trace?
We built Traces to share and learn how we're solving problems with agents.
Our team now adds traces.com links to every PR. This new context shows us how problems were approached.
It's a nice add-on to our workflow without disrupting it.
Traces is also a network. You can share privately, publicly, or via direct-link. All your public traces are available on your profile.
People are sharing all kinds of traces. Here's one of mine:
www.traces.com/s/jn7c679tvt...
Traces is a CLI app that can be downloaded from Homebrew, npm or curl.
When you run `traces`, you will see a list of your coding agent sessions. We support Cursor, OpenCode, Claude, Codex, Gemini CLI, Pi, with more adapters coming soon.
Get started at traces.com.
Introducing Traces
A new way to share and discover traces from coding agents, and a small step to make AI more multiplayer.
Here's how it works:
Thanks for the constant support Dwayne!
Personally I'm really excited about building in the open and everything we’ll learn through the process. Stay tuned.
More information on blog.market.dev.
Open source makes sense for us since so many of our customers are already active in open source communities, and are using market.dev to offer support, training and consulting packages for premium users.
More details on product bounties coming up soon too!
The hosted offering at market.dev now has simpler pricing plans designed to scale as you grow your business
Today, market.dev is open source
We're relaunching as an open source project under an MIT License, with a hosted offering
The goal is to give developers a combined toolkit for growing a services business - with built in landing pages, payments, CRM & contracts
damn I totally forgot to reply! I don’t have a preferred pattern for those quite yet tbh, most of my data pipeline is pretty linear and every step depends on the last (ie blocking) for now
Yeah I agree it has a lot in common :). DSPy lets me keep the deterministic parts as code, and the LLMs handle what they’re best at. IMO using system prompts for specifying what you want, how you want it can lead to sometimes brittle systems that may need change as models evolve.
This was super cool, thanks for sharing. Answered some big questions for me.
I'm wondering if you've heard of DSPy.ai? Feels like you'd appreciate it (also its Typescript cousin, axllm axllm.dev)
For more info on the PROGRAM series check out lu.ma/program
Event updates mostly on Twitter x.com/tarunsachdeva
This event is open to designers, developers and anyone who wants to build applications that use Toronto’s open data infrastructure to inform, connect and empower the city in new ways.
More event details soon!
Announcing PROGRAM: Toronto
A hackathon held at City Hall with the theme of making Toronto the world’s most programmable city.
PROGRAM: Toronto is June 20 @ 3pm, in collaboration with the City of Toronto Open Data Team, held on the eve of Toronto Tech Week.
Grab your ticket here: lu.ma/apcio9op
Anyways, sign up to market.dev and feel free to DM if this is something you'd want to try out!
Being able to have an agents ready for every workflow is one of the the unexpected design benefits of building this as a 'business in a box'
But we're not going to build these all ourselves, and the potential here is to turn into an MCP client so you can bring your own agent against your data