SekaiSync

SekaiSync skill

Use this skill whenever the conversation involves Project Sekai characters, songs, cards, events, units, stories, or fan-created content.

Workflow

1. Check sekaisync_freshness to confirm data coverage.
2. Resolve every proper noun with sekaisync_resolve_name.
3. Look up facts with sekaisync_lookup.
4. Request sekaisync_fact_pack for the entity IDs the answer needs.
5. Verify generated claims with sekaisync_verify_claims.

Output contract

• Use only the official localized name returned by SekaiSync.
• Do not add facts absent from the returned record.
• Treat official role labels in character profiles as snapshots of team responsibilities, not exhaustive capability or ownership lists. A label such as lyricist does not imply inability to compose or play an instrument, or lack of ownership of one. Require explicit supporting evidence for negative claims; missing evidence is unknown, not false.
• If there is no match, state that the local knowledge base does not cover the entity.
• The local store covers master metadata only. Full story text, images, audio, Live2D, charts, news and player data are marked missing in freshness; return unverified for those categories.
• Use sekaisync_web_lookup for crawled text from altsource_sv / altsource_ms. Crawling itself stays a CLI action with TOS consent.
• Base secondary creation on fact packs and avoid redistributing copyrighted assets.

术语刮削干预

当用户要求刮削/提取用语时，沿用原来的提取命令。程序返回 review_next 后，由正在执行任务的智能体自身继续导出、阅读、判断、提交，不需要用户另配 API Key。不要把程序返回待审队列当作整个刮削任务已经完成。

任务分为 discovery（直接从原文补找分词漏掉的用语）、translation（在对应的本地译文中找实际表面形）和 occurrence（判断具体出现及语境词义，不发布全局译名）。单故事用语会自动得到出现级任务，原有两故事通用译名门槛不降低。程序提供内容寻址的任务 ID、原文/译文语句、故事和位置。已完成的任务不会因 429 或重启而重复应用；下一次 export 会轮转到其他未决项。用户无需改用新命令。

# 1. 导出待裁决队列（自包含：每条含 term/候选/证据，无需回读语料）
sekaisync terms review export --out queue.txt --limit 20

# 2. 读 queue.txt，逐条给出判断：
#    decision: accept（采纳 hint）| reject（否定）| replace（补正确译名，需填 value）
#    每条写一句语境理由 rationale；有 task_context 的任务用 generalize: null
#    discovery 沿用 terms 数组；完整出现和非连续表达可用导出契约中的 subjects

# 3. 写回
sekaisync terms review submit --file judgments.txt

# 4. 检查 errors、applied_slots、remaining，再次 export 处理后续任务

# 查看沉淀与复用率
sekaisync terms review stats
sekaisync terms review methodology