agentleFS
superwyhsuperwyh/Game-Localization-Engineering/game-localization/SKILL.md

game-localization

Skill7 starsChanged 2 days agoMITView on GitHub

Localize game projects through text extraction, terminology and character guidance, translation and naturalness review, guarded resource integration, language QA, and detailed developer reports. Use for full projects, incremental updates, translation audits, or game translation post-editing; enable complex-script and compliance checks only for the requested scope.

3 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions3 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 3

    1. fcsouza/agent-skills★ 26same name
    2. jammyfu/open-game-skills★ 13same name
    3. pcmxiao/game-localization★ 1same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read workflow, translation and review, and reporting.

    line 15
  2. Distinctive:

    Deterministic checks: Check ID sets, source revisions, parameter types/bindings, tags, numbers/units, omissions, intentional empty strings, …

    line 37
  3. Very specific:

    Read the configuration contract, then adapt the example configuration.

    line 44

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Game Localization Engineering

Deliver traceable translations, importable resources, evidence of checks actually performed, and a detailed developer review report. All required execution rules, tools, and examples are included locally. Scripts use Python 3.10+ standard libraries without network access. The executing agent translates using project context. The project supplies its editor, compiler, fonts, and runtime.

English is the maintained primary edition. The documentation index provides installation and usage guides in 20 language and regional editions; detailed engineering references remain in English. The Chinese companion guide provides extended operating instructions. Documentation language does not select the game's target languages. Report language is also independent: set report_language to en (default) or zh-CN. Write decisions and review notes in that language; preserve game strings, IDs, and evidence.

Start and select modules

Read workflow, translation and review, and reporting. Inspect project instructions, the existing pipeline, and user edits. Record the authoritative source language/revision, selected language/script/region combinations, engine/package versions, resource types, and delivery scope.

Derive targets from the request and project; do not select every listed language. If direction or targets are missing, inventory first and clarify before bulk translation. Do not ask again for choices already provided. A text-only assignment does not authorize restructuring the localization framework.

TriggerReadWhen outside scope
Selected languages/regionsRelevant locale profilesDo not translate unselected languages
Translation drafting, review, or requests to remove translationese or an assistant-like voiceNaturalness and post-editing; use its examples for difficult decisionsKeep good text; do not infer authorship or optimize detector scores
RTL, shaping, dictionary segmentation, vertical text, special input, or actual UGC needsRelevant complex-script checksRecord exclusion; do not run a global script matrix
Extraction, integration, parameters, markupFormats and CLIPrefer a verified existing project adapter
Source/guidance updates or returned review tablesIncremental and review exchangeDo not treat returned rows as approvals
Archive/image/binary inventory or measured scenario coverageResource coverageKeep ordinary text-only work within its requested scope
Engine integrationRelevant engine branchDo not install unrelated engines
Delivery validationBaseline and triggered LQA casesExplain skipped checks; never label them passed
Requested compliance or specified market release reviewSelected regions/platforms/mechanics in complianceDefault off; language does not imply country
Conflicting, outdated, uncertain evidenceEvidence and decision rulesRecord unsupported conclusions as unresolved

Execution loop

  1. Baseline. Inventory IDs, file hashes, syntax, context, and audiovisual dependencies. Test an unchanged round trip, then a small loop with menus, parameterized dialogue, and a critical asset.
  2. Language assets. Build concept-based terminology and character × addressee × story-stage voice records. Separate UI, instructions, dialogue, subtitles, store copy, and legal text. Include examples, counterexamples, and exceptions.
  3. Translation and review. Work on complete messages, branches, and puzzle dependency groups. Establish source meaning, edit idiom/register, read as target-language text, then recheck facts, agents, negation, conditions, numbers, uncertainty, and game function against the source. Preserve intentional voice and clues. Record before/after and reasons for substantive edits; proofread and reconcile scenes. Resolve context gaps without inventing facts; continue independent entries. Distinguish self-review from native/developer review.
  4. Deterministic checks. Check ID sets, source revisions, parameter types/bindings, tags, numbers/units, omissions, intentional empty strings, and protected literals. Use real parsers for complex messages. Regex, counts, and glyph coverage do not prove syntax, layout, or runtime correctness.
  5. Staging and integration. Stage separately and inspect additions, changes, removals, and nontext fields. Continue authorized integration with recovery records. Stop affected overwrites and reconcile if source or destination changes. Translation does not itself authorize publication, rating submissions, or messages to others.
  6. Validation and repair. Perform applicable import, build, runtime, linguistic, accessibility, and enabled compliance checks. Retest fixes. Complete independent work when an environment is missing; report limits without inventing a pass.
  7. Developer report. On completion, partial completion, or stopping, deliver a detailed report, full source/target comparison, significant decisions, issues/evidence, unperformed checks, and specific developer actions. Delivery is incomplete without these.

Local tools

Read the configuration contract, then adapt the example configuration. SKILL, PROJECT, and RUN mean actual paths; quote them as required by the shell.

python -X utf8 SKILL/scripts/l10n.py inventory --project PROJECT --out inventory.json
python -X utf8 SKILL/scripts/l10n.py extract --config project.json --run RUN
python -X utf8 SKILL/scripts/l10n.py qa --run RUN
python -X utf8 SKILL/scripts/l10n.py stage --run RUN
python -X utf8 SKILL/scripts/l10n.py apply --run RUN
python -X utf8 SKILL/scripts/l10n.py report --run RUN

The CLI directly supports flat JSON maps and CSV/TSV. Other formats require native export or project adapters. Staging requires current review records for every selected entry, not every language or jurisdiction. Complete substantive review-notes.json content before the final report. Static checks, language review, integration, builds, runtime, and developer signoff are separate states.

For updates, use refresh --previous OLD_RUN --config project.json --run NEW_RUN; inspect stale entries, original translation bases and retained orphans. Bind relevant project terminology/voice files with guidance_files so their changes invalidate affected reviews. Use review-export/review-import for external CSV review, with truthful reviewer identity. Use coverage-init/coverage-check only when resource/scene accounting is needed. Their detailed contracts and examples are in the two references above.

Use the verification guide and example resources when first integrating. CLI stage, apply, rollback, refresh, review exchange and coverage-check attempt reporting after success or reportable failure; complete the human judgments and regenerate the final version. The reporting-only override report --run RUN --language zh-CN changes report presentation without changing the game or baseline.

Engineering invariants

  • Preserve internal keys, branch indices, control codes, asset references, parameter identities, story facts, and functional values. Translations are not identifiers.
  • Review regional adaptations as whole sentences. Software style guides do not define every character's voice. Do not infer gender or identity from names.
  • Scope compliance to applicable cases. Do not automatically remove plot, soften profanity, change mechanics, or claim approval. Separate recommendations from changes made.
  • Record fonts, translation, voice tracks, subtitles, format locale, and distribution regions separately. Fallback does not fix missing translation. Source/context changes invalidate old review.
  • Batch by scene/chapter and save progress/context until authorized work is complete. Length does not justify omitted branches; sampling is not full coverage.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.