legacy-tests

Files (1)

• SKILL.md

SKILL.md (first lines)

# Legacy tests

Legacy code is code without tests. Before changing it, capture what it does now — right or wrong — so any change in behavior is visible.

## 1. Pick the seam

- Identify the smallest unit you can call from a test with the behavior you need to protect: a function, a class, a CLI command, an HTTP endpoint.
- If nothing is callable in isolation, find a seam: a parameter you can pass, a dependency you can inject, a function you can extract _mechanically_ (no logic changes) to make it testable. Keep these enabling changes tiny and separate.
- Prefer testing at a higher level (endpoint, command) when internals are tangled; you can add finer tests later.

## 2. Characterize behavior

- Call the code with representative inputs and assert on the **actual** outputs you observe, including surprising ones. Name such tests clearly: `currently_returns_null_for_empty_input`.
- Cover: typical inputs, boundaries, invalid inputs, and error paths. Use real production-like examples (anonymised) when available.
- For outputs that are large or complex, snapshot/golden-file tests are acceptable — keep them readable and reviewed, and normalise volatile parts (timestamps, IDs, ordering).
- Pin external effects: what gets written, sent, or logged. Replace the network, clock and randomness with fakes at the boundary.

## 3. Check the tests can fail

Temporarily break the code (flip a condition, change a constant) and confirm a test fails. If nothing fails, the tests are not protecting that behavior. Revert the break.

## 4. Separate "is" from "should"

When you find behavior that looks like a bug, do **not** fix it in this step. Pin it, mark it (`// characterization: looks wrong, see issue`), and report it. Fixing it is a separate, deliberate behavior change with its own test.

## 5. Report

Which behaviors are now covered, how to run the tests, suspected bugs discovered, and what remains untested (and why — e.g. needs a real database).

From shankarr1503/omnistack, MIT.

More from shankarr1503/omnistack