Repository navigation
Conversation
Repeating the translator assignment makes the same workflow step require maintenance in two places. Keep the locale-rule obligation here without repeating the role invocation or leaving a dependent pronoun. The ordered translator/evaluator workflow and sequential-persona fallback remain at: https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/SKILL.md#L66-L68 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Maintaining a second evaluator assignment adds another copy of the same workflow duty. Keep the locale-specific review duties here while retaining the ordered review step and persona fallback in the process definition: https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/SKILL.md#L66-L68 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Restating score thresholds in the skill requires maintaining them alongside the mandatory locale rubrics. Keep the evaluator's rubric invocation and feedback, retry and escalation instructions; every supported locale already defines the same tier thresholds and all-applicable-criteria pass condition: https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/rules/es-es.md#L85-L92 https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/rules/ja-jp.md#L85-L92 https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/rules/ko-kr.md#L81-L88 https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/rules/pt-br.md#L85-L92 https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/rules/zh-cn.md#L85-L92 These are duplicate obligations, not evidence that the old thresholds were incorrect. The locale rubrics retain technical-content applicability, code-integrity exceptions, detailed scales, and their iteration cap. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Maintaining an abbreviated preservation checklist duplicates the detailed rules without their code-comment and frontmatter-value qualifications. The formatting section already requires the same Markdown, code, URL and frontmatter-key preservation and retains those qualifications: https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/SKILL.md#L102-L109 Remove only the duplicate checklist item; the complete rules stay intact. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The heading-link checklist duplicates a duty already maintained in the formatting section. That section retains localized-heading link updates and the exception for non-localized files and external targets: https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/SKILL.md#L111 Remove only the duplicate checklist item, not the duty or its exception. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Maintaining an abbreviated asset checklist repeats a duty whose detailed definition already accounts for the localized document's actual directory. Keep that definition, its root and nested-document examples, the localized asset exception and the real-file requirement: https://github.com/github/dev-days/blob/8572d1a0f69d5721836f000a9b20b30a067fc448/.github/skills/localizations/SKILL.md#L113 Remove only the duplicate checklist item. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
| ### Translator agent | ||
|
|
||
| Use the `translator` agent to perform the localization. It should follow the rules and guidelines defined for each target locale document in the `rules` directory. | ||
| The `translator` agent should follow the rules and guidelines defined for each target locale document in the `rules` directory. |
There was a problem hiding this comment.
🛠️ One translator assignment: Avoid maintaining a second invocation of the same role: the ordered translation workflow and persona fallback already assign this step. Naming the translator directly keeps the locale-rule duty clear after removing its duplicate antecedent. Commit
| ### Evaluator agent | ||
|
|
||
| Use the `evaluator` agent to assess the quality of the localized content. The evaluator checks for accuracy, cultural relevance, and overall quality, following the rules and guidelines defined for the target locale in the `rules` directory. | ||
| The evaluator checks for accuracy, cultural relevance, and overall quality, following the rules and guidelines defined for the target locale in the `rules` directory. |
There was a problem hiding this comment.
🛠️ One evaluator assignment: Avoid maintaining another copy of the same review step: the ordered workflow and persona fallback already assign evaluator review after translation. The locale-specific review duties remain here. Commit
| The evaluator checks for accuracy, cultural relevance, and overall quality, following the rules and guidelines defined for the target locale in the `rules` directory. | ||
|
|
||
| The evaluator scores the localized document against the locale's **Evaluator Scoring Rubric** (defined in the locale's `rules/<locale>.md`). The rubric uses two tiers: **Tier A** hard-fail criteria that must score 5, and **Tier B** graded criteria (1–5) that must score 4 or 5. A document passes only when **every applicable Tier A criterion scores 5 and every applicable Tier B criterion scores ≥ 4**. If anything falls short, return the document to the translator with specific notes and re-run the translate → evaluate loop until it passes, escalating to a human after the rubric's iteration cap. | ||
| The evaluator scores the localized document against the locale's **Evaluator Scoring Rubric** (defined in the locale's `rules/<locale>.md`). If anything falls short, return the document to the translator with specific notes and re-run the translate → evaluate loop until it passes, escalating to a human after the rubric's iteration cap. |
There was a problem hiding this comment.
🛠️ Rubric-owned thresholds: Avoid maintaining a second set of score thresholds: the mandatory Spanish, Japanese, Korean, Portuguese, and Chinese rubrics already define both the tier thresholds and the all-applicable-criteria gate. The rubric invocation and feedback, retry, and human-escalation instructions remain. Commit
| ## DOs and DON'Ts | ||
|
|
||
| - **Do** perform localization only for the target locales defined in the `rules` directory (one `<locale>.md` per supported locale). Do not localize into unsupported locales. | ||
| - **Do** preserve Markdown structure, code, external link/URL targets, and frontmatter keys exactly (see *Markdown and formatting preservation*). |
There was a problem hiding this comment.
🧹 Complete formatting guidance: Avoid maintaining an abbreviated copy of preservation duties without their qualifications. The formatting section already requires Markdown, code, URL, and frontmatter-key preservation while retaining the frontmatter-value and explanatory-comment exceptions. Commit
|
|
||
| - **Do** perform localization only for the target locales defined in the `rules` directory (one `<locale>.md` per supported locale). Do not localize into unsupported locales. | ||
| - **Do** preserve Markdown structure, code, external link/URL targets, and frontmatter keys exactly (see *Markdown and formatting preservation*). | ||
| - **Do** let heading anchors follow the localized heading text, and update same-document anchor links to match the new slugs so they keep resolving (see *Markdown and formatting preservation*). |
There was a problem hiding this comment.
🧹 One heading-link rule: Avoid maintaining the same heading-link duty in two places. The complete heading-anchor instructions already require localized-heading link updates and preserve the exception for non-localized files and external targets. Commit
| - **Do** perform localization only for the target locales defined in the `rules` directory (one `<locale>.md` per supported locale). Do not localize into unsupported locales. | ||
| - **Do** preserve Markdown structure, code, external link/URL targets, and frontmatter keys exactly (see *Markdown and formatting preservation*). | ||
| - **Do** let heading anchors follow the localized heading text, and update same-document anchor links to match the new slugs so they keep resolving (see *Markdown and formatting preservation*). | ||
| - **Do** point image and asset paths at the original assets (rewriting the relative path as needed so it resolves from `localizations/<locale>/`), unless a corresponding localized asset exists (see *Markdown and formatting preservation*). |
There was a problem hiding this comment.
🧹 Directory-aware asset guidance: Avoid maintaining an abbreviated asset-path rule alongside its complete instructions. The asset guidance and examples already account for the localized document's actual directory, allow corresponding localized assets, and require the target file to exist. Commit
There was a problem hiding this comment.
🟢 Approval recommended
The review records no active issues for this documentation-only consolidation, which removes duplication while retaining references to the relevant localization guidance.
No open issues 🎉
What changed in this PR
This pull request centralizes localization workflow guidance in the locale-specific rules and relevant formatting sections, reducing duplication in SKILL.md while preserving the complete instructions.
Changes:
- Simplifies translator and evaluator role descriptions.
- Removes duplicated Markdown, heading-anchor, and asset-path guidance.
- Delegates evaluator thresholds and iteration requirements to each locale rubric.
| File | Description |
|---|---|
.github/skills/localizations/SKILL.md |
Removes duplicated localization rules while preserving references to the full formatting guidance and locale-specific evaluator rubrics. |
Review effort: Lite (auto)
🧠 How this helps Copilot
Keep localization role assignments and formatting duties in one place so the skill does not maintain abbreviated copies alongside their complete instructions. The ordered translation and review process, code and frontmatter qualifications, heading-link exceptions, and directory-aware asset guidance remain intact.
Let each required locale rubric own its score thresholds instead of maintaining another copy in the skill. The evaluator still loads that rubric and follows its feedback, retry, and human-escalation requirements.
Automated details
Source:
curate-memoriesDream, analyzingmainat8572d1a.