Skip to content

README.md

Latest commit

 

History

History
82 lines (57 loc) · 2.28 KB

README.md

File metadata and controls

82 lines (57 loc) · 2.28 KB

types.generic-record-casts

Flags Record<string, unknown> casts on vague parsed/payload variables like data, payload, and parsed.

  • Family: types
  • Severity: strong
  • Scope: file
  • Requires: file.ast

How it works

The rule looks for as Record<string, unknown> casts assigned into generic object-bag variables such as:

  • parsed
  • payload
  • body
  • data
  • result
  • config

It gives extra detail when the cast comes directly from JSON.parse(...).

This pattern often shows up in generated or hurried glue code that turns unknown structured input into a generic property bag instead of validating it into a domain-shaped type.

To avoid obvious vendored noise, the rule skips very large bundled/generated files over 5000 logical lines.

Flagged examples

const parsed = JSON.parse(raw) as Record<string, unknown>;
const payload = response as Record<string, unknown>;
const data = value as Record<string, unknown>;

Usually ignored

const parsed = JSON.parse(raw) as UserConfig;
const token = value as { token: string };
const metadata = input as Map<string, string>;

How to fix / do this better

Treat unknown input as unknown for longer, then validate or narrow it at the boundary.

Better options:

  • parse into a real domain type with a schema or decoder
  • keep the value as unknown until you prove the fields you need
  • use a very local cast only when you immediately narrow and contain it
const input: unknown = JSON.parse(raw);
const parsed = UserConfigSchema.parse(input);

Or, without a schema library:

const input: unknown = JSON.parse(raw);
if (!isUserConfig(input)) {
  throw new Error("Invalid user config");
}

The goal is to avoid turning uncertain external data into a roaming generic object bag that downstream code has to keep guessing about.

Scoring

Each generic record cast adds 2 points. The file total is capped at 8.

Benchmark signal

Small pinned rule benchmark (manifest):

  • Signal rank: #7 of 9
  • Signal score: 0.67 / 1.00
  • Best separating metric: findings / file (0.75)
  • Hit rate: 3/6 AI repos vs 0/5 mature OSS repos
  • Full results: rule signal report