InterviewLab

AI agents

Real-world examples

Example: Add tests to existing code

Prompt examples are illustrative; commands are verified against official docs.

Scenario: a legacy module (src/reports/) ships without tests. Goal: discover the project’s test setup, generate tests that match existing conventions, run them, and iterate until green.

Workflow

  1. Discover the test framework, conventions, and how to run tests.
  2. Generate tests for one unit at a time, following existing test style.
  3. Run the tests; iterate on failures.
  4. Widen coverage incrementally.

Claude Code

claude

Discover the setup:

Figure out how this project runs tests: framework, config files, existing
test layout, and the exact command to run a single test file.

Generate tests that match house style:

Look at an existing well-tested module for style, then write tests for
src/reports/summary.ts covering the happy path and edge cases (empty input,
missing fields, date boundaries). Put them where this repo's tests live.

Run and iterate:

Run the new tests, fix any failures, and keep iterating until the file
passes. Then run the full suite to check for regressions.

Pre-approve just the test command so runs don’t prompt every time (permission rule syntax):

claude --allowedTools "Bash(npm test *)"

For unattended generation, print mode works too (no interactive session):

claude -p "Write unit tests for src/reports/summary.ts following the conventions in tests/, run them, and fix failures." --max-turns 30

Codex

Interactive, with the workspace-write sandbox so tests can run freely:

codex --sandbox workspace-write --ask-for-approval on-request
Inspect the repo's test setup: framework, config, directory conventions, and
the command to run one test file.
Following the style of the best existing test file, write unit tests for
src/reports/summary.ts (happy path + edge cases), run them, and fix
failures until green.

Scripted/CI variant — codex exec runs non-interactively (streams progress to stderr, prints the final message to stdout):

codex exec --sandbox workspace-write "Add unit tests for src/reports/summary.ts matching the conventions in existing tests. Run them and iterate until they pass."

Note: codex exec requires a Git repo by default (override with --skip-git-repo-check), and --full-auto is deprecated — use --sandbox workspace-write.

Kimi Code

kimi

Discover and generate:

Find how tests are organized and run in this repo (framework, config, test
file locations, single-file run command).
Write unit tests for src/reports/summary.ts following the conventions of the
existing test files: happy path plus edge cases for empty input, missing
fields, and date boundaries.

Run and iterate. Switch to Ask When Needed so repeated test runs don’t each prompt:

/yolo
Run the new tests and iterate until they pass, then run the whole suite to
catch regressions.

One-shot, fully non-interactive (kimi -p streams output to stdout and runs with auto permission — use only in trusted directories):

kimi -p "Add unit tests for src/reports/summary.ts matching existing test conventions, run them, and fix failures until green."

Last verified: 2026-09-07

AI coding tools evolve quickly. If you find outdated information, please open an issue or pull request.

View this doc on GitHub