Launched this week
Flatbread
Share the whole agent run, not a screenshot
64 followers
Share the whole agent run, not a screenshot
64 followers
Flatbread turns a coding agent's session into one link: what was asked, every step, what broke, what fixed it and the diff. Redaction runs on your machine before anything uploads. Another agent can pull the link and pick up where it ended.
The failed attempts are often the interesting part of an agent run. Sharing those alongside the diff gives a reviewer much more to work with than the final summary.
@oleksii_sekundant Agreed, the failures are usually where the useful stuff is. That's why the run page keeps what broke and what fixed it, not just the last message. And when another agent pulls the link, the brief starts with the blocker and what was already checked, so it doesn't go down the same dead ends.
The handoff brief is the part that matters most to me. A screenshot of the last message hides exactly what a second agent needs: what blocked the run and what was already checked, so it does not burn time down the same dead ends. When another agent pulls the link, does it get the diff and the file state along with that brief, or just the brief itself?
@jcbuilds Yes, it gets the diff too, not just the brief. The brief comes first (what blocked the run, what was checked, what fixed it), then the files that changed and their diffs, up to about 80 lines. If the change is bigger, it says how many lines were left out and points to the full diff on the run page. If a shell command made the edit, there's no diff, so it quotes the command instead.
It doesn't send a snapshot of the files, only what changed, and it never applies anything to your code. The second agent reads it as context and does the work in its own repo. And if the author shared the run without its code, the brief says the diff was withheld, so the agent knows something is missing.
"Another agent can pull the link and pick up where it ended" — okay that's the part I didn't know I needed.
@avinashbussa Thanks! That was the part I wanted most too. In a chat you tell the agent "pull flatbread " (or run flatbread pull <id>). It gets a short brief: what blocked the run, what was checked, and what fixed it, so it starts from there.
Does the redaction on the review page catch people's names and email addresses, or mostly keys and tokens? A lot of our agent sessions read CRM records, so names are what I'd worry about in an unlisted link