gosilent

Context-efficient test runner for AI agents. Wraps go test -json and produces ultra-compact output -- a single line for the entire passing suite, failure details only when needed. Designed to minimize LLM context token consumption.

Why?

go test output grows with every package. A 10-package run produces 10 lines (one per package). When an AI agent runs tests in a loop, every token of that output counts against its context window.

Common workarounds pipe through grep or tail, then re-run go test on failure to get details. That doubles execution time.

gosilent solves both problems: it parses the full JSON stream in a single run, emits one line for the entire suite on success, and expands only the failures.

Before: go test ./...

ok  	github.com/stretchr/testify/assert	0.42s
ok  	github.com/stretchr/testify/http	0.31s
ok  	github.com/stretchr/testify/mock	1.20s
ok  	github.com/stretchr/testify/require	0.38s
ok  	github.com/stretchr/testify/suite	0.81s
?   	github.com/stretchr/testify	[no test files]
?   	github.com/stretchr/testify/...	[no test files]

After: gosilent test ./...

ok  all packages (739 tests, 3.12s)

7 lines collapsed to 1.

Output Format

# All tests pass -- single line, any number of packages
ok  all packages (67 tests, 1.71s)

# Single package -- shows package name
ok  formatter (21 tests, 0.00s)

# Significant flags are shown
ok  all packages -race (67 tests, 7.14s)

# Test failures -- only failing packages expanded
FAIL  example.com/foo
  TestSub
    sub_test.go:15: expected 3, got 5

1 failed, 2 passed (67 tests, 2.43s)

# Build failure
FAIL  example.com/broken [build failed]
  foo.go:10:5: undefined: DoesNotExist

1 failed (0 tests, 0.00s)

No-test-files packages are completely omitted. Skipped tests within passing packages are not mentioned. Per-package pass lines are eliminated.

Installation

go install github.com/AntiD2ta/gosilent/cmd/gosilent@latest

Or build from source:

git clone https://github.com/AntiD2ta/gosilent.git
cd gosilent
make build

Requires Go 1.26+.

Usage

# Run all tests
gosilent test ./...

# With go test flags (passed through)
gosilent test -race ./...
gosilent test -run TestFoo ./pkg/

# Per-package detail mode
gosilent test --detail ./...

# Verbose mode -- bypass JSON parsing, raw go test output
gosilent test --verbose ./...

# Print version
gosilent version

gosilent test [args...] runs go test -json [args...] under the hood. The -json flag is injected automatically unless already present or --verbose is set.

Flags

Flag	Description
--detail	Per-package output with test counts (e.g., PASS pkg 42/42 3.45s)
--verbose	Raw go test output, bypasses JSON parsing entirely

--verbose takes precedence over --detail since it bypasses JSON parsing. Everything else passes through to go test unchanged.

Benchmarks

gosilent vs go test on open-source projects (tokens at ~4 chars/token):

Project	go test	gosilent	Ratio
cockroachdb/pebble (11,561 tests)	34,783 tokens	12 tokens	2,899x
kubernetes/client-go (1,872 tests)	6,222 tokens	12 tokens	519x
prometheus/prometheus (9,413 tests)	9,774 tokens	37 tokens	264x
kubernetes/api (1,812 tests)	735 tokens	11 tokens	67x
stretchr/testify (739 tests)	103 tokens	11 tokens	9x

Total across 11 projects: 51,952 → 145 tokens (358x reduction).

Reproduce with ./scripts/benchmark.sh.

How It Works

gosilent spawns go test -json as a subprocess and streams its output through a JSON parser that builds per-package results. A compact formatter then aggregates all packages into a single summary line on success, or expands only the failing packages with their test names and assertion output -- stripping Go's test boilerplate (=== RUN, --- FAIL, etc.).

Dependencies

• urfave/cli/v2 -- CLI framework
• stretchr/testify -- test assertions (dev only)

License

Apache License 2.0