Localization Resource QA

Check the resource contract before testing the interface.

Purchases are not open yet. Explore the tools and their limits below.

For small, explicit catalogs

For developers with flat JSON resources and project-owned rules. Check source and target resources without rewriting translations.

Find missing keys and simple placeholder-count mismatches.
Check plural-category coverage and direction metadata against the supplied contract.
Separate confirmed contract errors from extra-key, unchanged-text, length and bidi-control review flags.

What is included

Portable SKILL.md instructions and Python auditor
Input-contract and report references
Synthetic before/after examples
33 regression tests and a separate practice task

The package includes synthetic examples and regression tests. These are not production accuracy or translation-quality evidence.

Three inputs → one JSON report

Supply a source resource, target resource and explicit project contract. Each UTF-8 JSON input is limited to 5 MiB. Resource objects have exactly locale, direction, strings and plurals fields.

python3 localization-resource-qa/scripts/audit.py \
  --source "source.json" --target "target.json" \
  --contract "contract.json"

Fixture excerpt

Source: "welcome": "Hello {name}"
Target: "welcome": "Hallo"

Contract requires direction: ltr
Target declares direction: rtl

These produce PLACEHOLDER and DIRECTION_METADATA errors. Other defects in this fixture contribute to the total.

Source · Target · Contract

Actual example result

{
  "status": "fail",
  "counts": {
    "error": 5,
    "review": 2
  }
}

Errors include a missing key and missing contracted plural category. Review flags identify an extra key and an advisory length concern.

Read the complete synthetic report

Strict input. Narrow conclusions.

Only plain simple {identifier} tokens are supported; token counts must match. The project owner supplies plural and direction rules. They are not guessed from locale names.
Not generic framework JSON: nested catalogs, ICU, printf, markup, escaped braces and framework syntax require a different parser or a separately reviewed adapter, which is not included.
Unfamiliar or unsupported input is rejected. It must never be treated as a passed audit.
No native-language review, cultural assessment, runtime plural-number selection, actual RTL rendering, clipping, security or accessibility certification.
Code-point length is advisory, not pixels. Bidi controls can be legitimate; findings are not instructions to delete them.

Quick questions

Can this certify a translated release?

No. A zero-finding report only covers these technical checks and the supplied contract. Human language review and application testing remain necessary.

Does exit 0 always mean no findings?

No. Both pass and review statuses exit 0. Inspect status and review counts. Errors exit 1; invalid input exits 2.

Does it translate or repair files?

No. It reports issues without rewriting source or target files. Included translations are synthetic test data, not native-language exemplars.

Runtime & data handling

The scripts use Python 3.10+ features and only the standard library. Actual script testing: Python 3.12.14 on Linux. Other runtimes and operating systems are unverified.

Included SKILL.md instructions can guide an agent to run the script. Installation and automatic discovery in external hosts, including Claude and other Codex clients, have not been verified.

The helper processes files locally without network calls or source-file edits. This is not an end-to-end privacy promise: your agent host may transmit text to its model provider. Review its configuration before supplying confidential files.

Availability

We are preparing these tools for sale. Purchases and payment links are not available yet. No release date is promised.

Final prices, commercial licensing, refund terms, support and delivery details will be shown before any purchase is offered. No purchase license or service commitment is granted by this page.