Compact cargo output

A long cargo build or cargo test can hand an agent thousands of tokens of compiling lines, warnings, and progress bars - enough that reading a single task output can eat most of a session's context window. banish compacts that output before it reaches Claude Code, Cursor, or any MCP agent, keeping the errors and the final status and dropping the rest.

The problem - cargo output is verbose

cargo build prints a Compiling line for every crate in the graph, then a stream of warnings, then the final result. On a cold build of a real project that is hundreds of lines the model does not need in order to know what happened. When an agent reads that from a background task, the whole log can be pulled into context in one shot.

Before and after with banish

banish runs the real command and pipes the output through the cargo filter:

The filter drops the per-crate Compiling and Downloading lines and keeps the last few lines - the Finished line on success, or the error summary when the build fails. See the full spread across commands on the benchmarks page, measured on a fixed corpus you can reproduce with banish bench.

How the cargo filter works

The cargo filter ships with banish out of the box. It compacts cargo build, cargo test, and cargo clippy. On cargo test it keeps the lines that matter - errors, warnings, the test result, and panics - and drops the passing noise. If a filter fails, banish returns the raw output rather than swallowing it. See built-in filters for the full set.

Write your own for another command

The same approach works for any noisy command. The .bsh language lets you define a filter in about ten lines - no recompile. To start compacting cargo output in your own agent, set up an agent.