~/api main ❯ cargo test
Compiling api v0.4.2 (/Users/dev/api)
Finished `test` profile in 4.82s
Running unittests src/lib.rs
running 212 tests
test auth::login ... ok
test auth::logout ... ok
test auth::refresh ... ok
test billing::invoice ... ok
test keep::attach ... FAILED
test keep::detach ... ok
test keep::replay ... FAILED
test sessions::discover ... ok
… 199 more lines …
test result: FAILED. 210 passed; 2 failed
Everything, byte for byte — 212 lines land in the agent’s context.
~/api main ❯ compact cargo test
test result: FAILED. 210 passed; 2 failed; finished in 4.82s
✗ keep::attach — assertion failed: replay.len() > 0
✗ keep::replay — broken pipe talking to broker
[210 passing lines elided · raw → trace://cmd_42/stdout]
The cargo-aware compactor keeps failures and the verdict — 4 lines.
~/api main ❯ semantic cargo test
{
"command": "cargo test",
"exit_code": 1,
"status": "failed",
"headline": "exit 1: 2 failure-like line(s)",
"failures": ["test keep::attach ... FAILED", "…"],
"raw_stdout": "trace://cmd_42/stdout"
}
A structured observation an agent can parse — with the raw bytes one trace:// away.
~/api main ❯ lossless-ref cargo test
[exit 1 · 212 lines · 14.8 KB] trace://cmd_42/stdout
One line now, every byte later — agtrace or cat the reference on demand.
~/api main ❯ silent cargo test
~/api main ❯
Exit code only — the prompt turns red, the output waits in the trace store.