Testbench
Convention-aware testing
Tests that match your project's style, not generic AI output. Testbench reads your existing tests before generating anything — extracting import style, mock patterns, assertions, and naming conventions. Generated tests are indistinguishable from hand-written ones. Self-healing: runs, fixes, and retries up to 3 times.
Skills
/testbench:calibrateTest Framework Calibration
Detects test framework (Vitest, Jest, Playwright, Cypress, pytest, Go test, Cargo test, RSpec, PHPUnit). Reads conventions. Generates config.
- Queries Context7 for framework docs
- Extracts conventions from existing test files automatically
/testbench:generateConvention-Aware Generation
Generates tests for a single file. Reads config AND existing test files, never generates without context.
- Categories: happy path → edge cases → error cases → type contracts
- React: render, content, interactions, conditional rendering, accessibility
- API routes: successful response, validation, auth, authz, not found
- Auto-splits if exceeding 150 lines
- Runs test and fixes up to 3 times before declaring failure
/testbench:runTest Runner
Runs tests with scope control: all, changed files only, or single-file. Parses output, extracts failures with source context.
- Supports Vitest/Jest/pytest/Go/Cargo output formats
- Coverage analysis with Composure graph integration
- Identifies high-impact untested functions by caller count