Golden testing for CLI applications - a TypeScript port of trycmd.
Note
100% of the code and specs in this repository were written by Claude Code. The design and management and prompting was by me (jlevy) supported by the workflows, agent rules, and other research docs in Speculate.
You can see what you think, but I find the code quality higher than most agent-written code I've seen because of the spec-driven process. You can review the architecture doc and all of the specs all of the specs in docs/project. The general research, guideline, and rules docs I use are in docs/general.
Write CLI tests as Markdown. tryscript runs commands, captures output, and compares against expected results. Tests become documentation; documentation becomes tests.
---env: NO_COLOR: "1"sandbox: true---# Test: CLI help```console $ my-cli --helpUsage: my-cli [options] <command>Options: --version Show version --help Show this help...? 0```# Test: Version output```console $ my-cli --versionmy-cli v[..]? 0```# Test: Error handling```console $ my-cli unknown-command 2>&1Error: unknown command 'unknown-command'? 1```# Test: Check output file contents```console $ my-cli process data.json > output.txt && grep "success" output.txt[..]success[..]? 0```The [..] matches any text on that line. The ... matches zero or more lines. These "elision patterns" let tests handle dynamic output gracefully. Any shell command works - pipes, redirects, environment variables, etc.
# Install pnpm add -D tryscript # For coverage support (optional) pnpm add -D c8 # For accurate line counts when merging with vitest (optional) pnpm add -D c8 monocart-coverage-reports # Run tests npx tryscript run tests/ # Update expected output when behavior changes npx tryscript run --update tests/- Markdown format - Tests are readable documentation
- Elision patterns - Handle variable output:
[..],...,[CWD],[ROOT],[EXE] - Custom patterns - Define regex patterns for timestamps, versions, UUIDs
- Update mode - Regenerate expected output with
--update - Sandbox mode - Isolate tests in temp directories
- Code coverage - Track coverage from subprocess execution with
--coverage(experimental; use--coverage-monocartfor best accuracy)
tryscript run [files...] # Run golden tests tryscript coverage <commands...># Run commands with merged coverage tryscript docs # Show syntax quick reference tryscript readme # Show this documentation tryscript --help # Show all optionsFor complete syntax reference, run tryscript docs or see the reference documentation.
| Option | Description |
|---|---|
--update | Update test files with actual output |
--fail-fast | Stop on first failure |
--filter <regex> | Filter tests by name |
--verbose | Show detailed output |
--coverage | Collect code coverage (requires c8) |
--coverage-monocart | Use monocart for accurate line counts (requires monocart-coverage-reports) |
--coverage-exclude-node-modules | Exclude node_modules from coverage (default: true) |
--coverage-exclude <pattern> | Exclude patterns from coverage |
Note: Coverage features are experimental. See the reference documentation for details on merged coverage, monocart integration, and sourcemap requirements.
# Clone and install git clone https://github.com/jlevy/tryscript.git cd tryscript pnpm install # Build and test pnpm build pnpm testMIT