SignalDrift/Crow-public ? reverse-engineered prompt
Reverse engineered prompt
Build out this Crow project so it works as a small Python control plane for coordinating autonomous engineering agents.
I want it to manage bounded work, claims, evidence, review, and safe continuation using durable records, with the state always coming from the repository data instead of model memory. Make the worker, reviewer, reconciler, and status tools actually usable, and keep the behavior fail closed whenever authority is unclear or a transition is not allowed.
Please follow the README and the protocol docs, use the existing tests as guidance, and make sure the command line tools can parse records, reduce state, and show a clear read only status. If anything is missing, infer the simplest sensible implementation from the docs and current tree, and look up current docs online if you need to. Also make sure the examples stay free of secrets and machine specific values.
Are you gonna build this?
make sure you review the code using coderabbit