Abbreviatory Launches With Run History, an MCP Server and Git Diff Reports

Abbreviatory, the studio’s twenty-first project, deals with output nobody reads in full: a test run with 1,204 passes and 7 failures, a 5.6 MB log, a 720 kB API response. It cuts that down to a short report by rules, with no language model. Failures, errors and warnings come first, repeats fold into one line with a count, and the report never goes over a byte budget. Everything it leaves out has a reference that brings back the original lines, byte for byte. That log of 63,369 lines comes out at 3,075 bytes.

It launched on 18 October and moved fast. 0.2.0 added run history: new failures, fixed ones, how long a test has been failing, which ones are flaky, and one line when nothing changed. 0.3.0 made it an MCP server, so a coding agent can run its tests through it and read the short report, plus an HTTP service and a browser demo running the engine as WebAssembly. 0.4.0 added Git diffs, curl responses and directory listings to go test, pytest, JUnit, logs and JSON.

The testing holds it to four promises on every fixture at every budget from 256 bytes to 16 kB: never over budget, every failure shown or counted, the same bytes for the same input, and every reference expanding back exactly. Run history was checked on 1,500 simulated CI runs, with every reported change matching the known one. Each release went through an independent review first, and every finding has its own test. It’s one static Go binary, open source under Apache 2.0, with binaries for Linux, macOS and Windows. jvq also runs in CI, as a gate on benchmark results; Abbreviatory makes the test output itself short enough to read.