Coding agents are fast. Reconstructing what the team meant is still slow.
HoldUp is built around one simple idea: before an agent writes code, give it the few product and engineering constraints that actually matter for that task.
The problem
A real engineering task rarely lives in one place. The ticket has the request. An ADR explains the architecture. An API schema defines compatibility. A migration doc contains rollout rules. Another repo has a consumer. A past pull request explains why the weird code exists.
A senior engineer often knows how to connect those pieces. A coding agent sees whatever context we remembered to give it.
What we are building
HoldUp starts with one task and creates a work packet: the goal, the relevant constraints, things that should not change, evidence for each constraint, and any questions that need a human decision before implementation.
The resolved packet goes to the coding agent. When the pull request opens, HoldUp checks the implementation against the same packet.
Why now
Coding agents are making implementation dramatically faster and increasing the amount of software that can be produced. That makes task understanding, compatibility, rollout context, and review intent more important rather than less important.
Current stage
HoldUp is early and working with design partners. We are deliberately starting with individual real tasks and failures instead of asking teams to migrate their process or connect their entire company on day one.
Who is building it
HoldUp is being built independently by Vraj Patel, a software engineer and CS master's student at the University of Maryland.
Have a task your agent misunderstood?
Send us the ticket or PR. We will reconstruct the context it should have received and show you the HoldUp work packet.
Try it on one task