OpenParable
Skip to content
← Back to skills

Compat Review

Testing & Quality
compat-review

Compat Review is an Agent Skill for testing & quality published in the toss/es-toolkit repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Verify compat PR claims by running lodash vs es-toolkit/compat at runtime

View on GitHub ↗
QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

  1. Open your project in Codex.
  2. Copy and paste the prompt below.
  3. Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/toss/es-toolkit/blob/HEAD/.claude/skills/compat-review/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/compat-review, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

Compat Layer Review

Verify that a compat PR actually fixes a real lodash inconsistency.

Input

$ARGUMENTS — PR number (e.g. 1234) or function name (e.g. chunk)

Workflow

1. Identify target function and PR claims

PR number:

gh pr view {number} --repo toss/es-toolkit --json title,body,files

From the PR description and diff, extract:

  • Which compat function is being fixed
  • What inconsistency the contributor claims (expected vs actual behavior)
  • Any test examples the contributor provides

Function name: Read the function source and its spec to understand current behavior.

2. Build comparison test

Create a temporary vitest spec at src/compat/{category}/_compat-review-{fn}.spec.ts.

import { describe, expect, it } from 'vitest';
import { fn as compatFn } from 'es-toolkit/compat';
import { fn as lodashFn } from 'lodash';

Include two groups of test cases:

A. Contributor's claimed examples — Extract directly from the PR description or test diff. These are the cases the PR claims to fix.

B. ~10 additional edge cases you generate — Based on the function's signature and lodash's known behavior patterns:

  • Empty inputs: [], {}, '', 0
  • Nullish: null, undefined
  • Negative/zero/float numbers: -1, 0, 1.5, NaN, Infinity
  • Type coercion: string numbers ('3'), boolean, mixed types
  • Boundary: single-element arrays, very long strings, deeply nested objects
  • Pick cases that are relevant to the specific function

Each test:

it('description', () => {
  let lodashResult, compatResult;
  let lodashErr: any, compatErr: any;
  try {
    lodashResult = lodashFn(args);
  } catch (e) {
    lodashErr = e;
  }
  try {
    compatResult = compatFn(args);
  } catch (e) {
    compatErr = e;
  }

  if (lodashErr && compatErr) return; // both throw = consistent
  if (lodashErr || compatErr) {
    throw new Error(`Behavior mismatch: ${lodashErr ? 'lodash throws' : 'compat throws'}`);
  }
  expect(compatResult).toEqual(lodashResult);
});

3. Run comparison BEFORE the PR change

yarn vitest run src/compat/{category}/_compat-review-{fn}.spec.ts

Record which tests fail — these are real inconsistencies that exist on main.

4. Apply PR changes and re-run

gh pr diff {number} --repo toss/es-toolkit | git apply

Run the same test again. Confirm that:

  • The contributor's claimed examples now pass
  • The additional edge cases still pass (no regressions)

Revert after testing:

git checkout -- .

5. Clean up

Delete _compat-review-*.spec.ts.

6. Report

## {fn} — Compat Review

### PR Claim
{What the contributor says they're fixing}

### Verification (BEFORE fix)
| # | Input | lodash | compat | Match? | Source |
|---|-------|--------|--------|--------|--------|
| 1 | ...   | ...    | ...    | MISMATCH | PR example |
| 2 | ...   | ...    | ...    | MATCH    | Edge case |

### Verification (AFTER fix)
| # | Input | lodash | compat | Match? | Source |
|---|-------|--------|--------|--------|--------|

### Verdict
- [ ] PR claim is valid (inconsistency confirmed on main)
- [ ] Fix resolves the claimed inconsistency
- [ ] No regressions in edge cases
- [ ] Existing tests still pass