When to Use This Skill
- Use only when the user explicitly names CrossFrame Suite,
crossframe-suite,/crossframe-suite, or$crossframe-suite. - Use as the umbrella router when a CrossFrame task may need multiple sibling skills, such as diagnosis plus public analysis, essay output, and review.
- Do not load every sibling skill by default; follow the routed workflow and progressive-reading rules in the original body below.
Packaged Source Note
This AAS-ready copy preserves the original CrossFrame skill body below. Chinese remains the canonical semantic layer; English metadata is only for discovery, installation, and repository review.
Limitations
- The skill body is intentionally Chinese-canonical; English metadata is for discovery and does not replace the original Chinese terms.
- Use only after explicit CrossFrame invocation or
crossframe-suiterouting; do not apply it as a generic default reasoning layer. - It structures analysis, drafting, and review, but does not replace source verification, domain expertise, or legal, medical, or financial judgment.
crossframe-suite is the master scheduling skill and does not replace any specialized skill. It does only three things:
- Determine which workflow the user's task belongs to.
- Arrange the continuous reading order.
- Output a brief scheduling outline, then enter the corresponding skill.
It does not copy crossframe、crossframe-essay or the body text of other parallel skills. Chinese is the authoritative semantics; English is only used for skill ids, file names, and external communication names.
When a task triggers the CrossFrame main body, the suite must also include ../crossframe/references/runtime-read-policy.md and ../crossframe/references/continuity-closure-map.md in the scheduling judgment: whether this time requires continuous section co-reading according to the v5.0 original text, rather than reading only a single concept card. When package descriptions or source anchors are needed, downstream then reads continuity-bundles.md or specific package files in a targeted manner. v3.0 and v2.0 files serve only as historical baselines; by default, the v5.0 source structure is authoritative.
The interactive choices at the suite entry confirm only two things: output mode and role. Article type is not selected at the beginning of the suite; if it ultimately enters crossframe-essay and the article layer is not closed, first complete the problem decomposition and structural insight draft, then crossframe-essay/templates/article-type-selection-dialog.md confirms it separately before body text generation.
Master entry after explicit invocation
Only when the user explicitly invokes crossframe-suite、$crossframe-suite、/crossframe-suite or explicitly requests use of CrossFrame Suite does entry proceed from this skill. After entry, scheduling is prioritized by this skill; only when the task is very singular and the user names a specialized skill should the corresponding specialized skill be used directly.
When the user submits any CrossFrame content task through this skill as the master entry, by default the final output is always5.0 hybrid long-form article, with the output tier being full-visible-v5-longform: fully visible draft + complete long-form article body. Specialized artifacts may be generated first, but by default the process finally enters the article layer.
crossframe -> [needed sibling skills] -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
This default is not meant to make every answer long, but because an article is better suited to hand structural judgments, specialized artifacts, and v5.0 fidelity checks to ordinary readers: first complete the necessary diagnosis/review/case/teaching/debate/memo, then form the draft, then write the article, and finally pass the quality gate.
The complete interaction order is fixed as:/crossframe-suite -> 模式/角色选择器(4+6) -> suite 路由与专项拆解 -> 结构洞察底稿 -> 文章类型选择器 -> 写作技法读取 -> 文章正文 -> 质量闸收束. The article type selector must occur after the structural insight draft and before the article body.
This default does not directly pass a fixed voice to crossframe-essay. The voice is jointly determined by the role, output mode, and references/output-mode-selector.md in topic_sensitivity : academic expert/critical reflective thinker defaults to a neutral analytical style, while mass communication/future explorer may enable an editorial undertone; when the user explicitly requests "friendly/editorial tone/reply style," that overrides. Neutral analytical style is not cold style;vulnerable themes must still first connect with people.
full-visible-v5-longform means: the v5.0 continuous co-reading package, source structure fidelity, concept risks, and reverse conditions must be visible in the draft; but these backend checks must not swallow the body text. The body text must still be written as a complete article, with a title, development, conceptual ascent, return to reality, boundaries, and aftertaste.
crossframe-review is a quality gate, not the final writer of the default article-writing chain. As long as the article layer is not closed, the final visible deliverable must still include # 结构洞察底稿 and # 文章正文; when the quality gate passes, only a very short conclusion or internal pass is appended, and it must not output only a review report. Only when the user explicitly requests "only review/complete review report/no article," or when the quality gate finds a hard failure and publication must be blocked, is it permitted to make the review report the main output.
Only when the user explicitly says "only want/don't want an article/don't want it written up/short answer/three sentences/table/list/raw review/raw case library/raw memo/pure diagnosis/action plan only" should the default article layer be turned off. In that case, the deliverable specified by the user should be retained.
When to use
When the task is not a single diagnosis but may require multiple CrossFrame skills to collaborate in sequence, prefer using this skill:
- The user wants to use CrossFrame but has not specified a particular sub-skill.
- The user only says "analyze this/what do you think/talk about it/write something," and it looks more like they want to see a readable output.
- Writing articles, commentaries, thought pieces, public commentary, organizational retrospective articles.
- Answering readers' questions, editorial replies, consultative responses, but there is structural diagnosis behind the question.
- Organizing materials into cases, then writing analysis or distilling concepts.
- Reading, theory, and article research notes that need comparison with CrossFrame's connections and differences.
- After propositional debate, needing to write it up, give a conclusion, or write a rebuttal.
- First generate an output, then review whether it actually reasons.
- The user says "these skills should be used together," "trigger in sequence," "general rules," "general entry point," "how to combine and invoke them."
If the task is very single-purpose, directly use the corresponding specialized skill; do not take a detour:
- If only structural diagnosis is needed:
../crossframe/SKILL.md - If only an article is needed:
../crossframe-essay/SKILL.md - If only review is needed:
../crossframe-review/SKILL.md - If only a short reply is needed:
../crossframe-dialogue/SKILL.md - If only a case library is needed:
../crossframe-casebook/SKILL.md - If only the evidence boundary for public issues is needed:
../crossframe-public/SKILL.md - If only an organizational repair memo is needed:
../crossframe-org/SKILL.md - If only concept teaching is needed:
../crossframe-teach/SKILL.md - If only propositional argumentation is needed:
../crossframe-debate/SKILL.md - If only reading research notes are needed:
../crossframe-notebook/SKILL.md
Must read
Read after each trigger:
references/output-mode-selector.mdreferences/workflow-routing-map.mdprotocols/suite-dispatch-protocol.mdtemplates/suite-reasoning-outline.md
Then read the corresponding sibling skill according to the route. For basic structural judgment, usually first read ../crossframe/SKILL.md and ../crossframe/references/read-routing-map.md。
Scheduling principles
- Basics first: most complex tasks first use
crossframeto establish factual boundaries, scale windows, mechanism candidates, and judgment tiers. - Scenario append: read only the specialized skills necessary for this instance; do not trigger all skills together.
- Writing comes later: before writing an article, first have a structural insight draft; specialized judgments such as public, organizational, debate, and reading should be completed first, then enter
crossframe-essay。 - Default writing: when the suite is triggered, the final output by default goes through
crossframe-essay, and the output tier is fixed asfull-visible-v5-longform; specialized artifacts come first, and the article comes later. - Mode/role first: at the start of the suite, only confirm the output mode and role; when there is no trigger word, display
templates/mode-selection-dialog.mdand wait for a reply, without starting directly. - Article type comes later: the article type is confirmed only after entering
crossframe-essayand after the structural insight draft is generated; it determines the article's expressive form and the reading of writing techniques, and does not change the suite's main route. - Capsule attribution: the suite only passes in
selection_state、workflow_state、voice_modeand the article layer switch; it must not read the v5 source index, must not expand continuous-reading packages, and must not generatev5-read-state-capsule. Capsules are generated bycrossframethe core layer after hitting source modules, entry packages, and the must-read-together closure. - Voice is determined by role: when the suite writes by default, according to
output-mode-selector.md, passvoice_modeandtopic_sensitivitytocrossframe-essay. When the user explicitly requests "friendly/editorial tone/reply style," override. - Long-form contract: Any CrossFrame content task entered from the suite and not explicitly turning off the article layer is by default not a short answer, and must not replace the full body text with bullet-point diagnosis, summary-style answers, or "if you only want one sentence".
- Composition boundary: By default, all CrossFrame content tasks are composed; only when the user explicitly turns off the article layer with words such as "only/ no article/ short answer/ table/ checklist/ diagnosis only/ action plan only" is the article layer turned off.
- Source continuity: For high-responsibility, public institutions, intimate relationships, long-term evolution, deep analysis, framework governance, AI reality verification, weak signals/opacity, no exit, and article output, the triggered v5.0 continuous co-reading package must be listed in the dispatch; do not list only concept cards.
- Source anchor integrity: Whenever entering the article layer, high-responsibility, public institutions, AI/procedural artifacts, lifecycle, no-exit subjects, or framework governance, the dispatch outline must require downstream reuse
v5-read-state-capsuleand perform the source anchor integrity check. - Review closure: Important outputs by default finally use
crossframe-reviewas a quality gate; the quality gate must not take over the final output or swallow the draft/body text. Lightweight short answers may only perform internal self-checks. - Source-check restraint: Public issues, real institutions, platforms, policies, people, companies, and the latest facts must be source-checked; private relationships, philosophical generalities, and user-supplied materials by default are not source-checked.
- Plain language first: The final output first gives a result that ordinary people can understand, with terminology used only for necessary mapping.
Default continuous chain
结构诊断:
crossframe -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
开放式可读分析:
crossframe -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
普通洞察文章:
crossframe -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
公共评论文章:
crossframe -> crossframe-public -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
组织复盘/修复文章:
crossframe -> crossframe-org -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
答读者问/编辑回信:
crossframe -> crossframe-dialogue -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
案例沉淀:
crossframe -> crossframe-casebook -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
概念教学:
crossframe -> crossframe-teach -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
命题辩论:
crossframe -> crossframe-debate -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
辩论后成文:
crossframe -> crossframe-debate -> crossframe-essay -> crossframe-review
读书/理论研究:
crossframe -> crossframe-notebook -> crossframe-essay(full-visible-v5-longform) -> crossframe-review
读书后成文:
crossframe -> crossframe-notebook -> crossframe-essay -> crossframe-review
After entering crossframe-essay , first generate a structural insight draft, then execute the article type selection rule: when the user has explicitly specified the article type, record it in the draft and adopt it directly; when the user has not specified it and the article layer is on, display based on the draft ../crossframe-essay/templates/article-type-selection-dialog.md; when the user replies "default/ automatic/ either is fine", adopt the draft's recommendation; when the user explicitly turns off the article layer, do not display it.
crossframe-review-lite indicates that a full review report need not be output, but it must check: whether factual boundaries were skipped, whether there is personality judgment, whether there is concept stacking, and whether the evidence tier was exceeded.
Output method
Unless the user only asks "which skill should be used", the final output first gives a short dispatch outline:
调度提纲
- 任务类型:
- 输出模式与角色:
- 工作流:
- 必读 skill:
- 按需读取:
- 连续联读包:
- 主题敏感度:
- 正文声口:
- 文章类型:
- 输出档位:
- 不读取:
- 质量闸:
Then enter the corresponding skill's normal output. Do not make the dispatch outline longer than the task itself.
Prohibitions
- Do not trigger all skills just to appear complete.
- Do not skip
crossframe's factual boundaries and judgment tiers and write the article directly. - Do not fail to source-check public issues yet make strong judgments.
- Do not treat
crossframe-reviewas a formal ending; it must be able to veto bad output. - Do not, in the suite's default composition chain, output only the quality gate, review conclusion, or repair suggestions while hiding
结构洞察底稿and文章正文。 - Do not let dispatch rules replace the protocols of specialized skills.