crossframe-essay

CrossFrame Essay

When to Use This Skill

Use only after explicit CrossFrame Essay invocation or after crossframe-suite routes a CrossFrame task into article output.
Use for Chinese critical insight essays, public commentary, concept essays, long-form reader replies, and structure-to-article drafting.
Do not use as a generic writing skill outside explicit CrossFrame context.

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-suite routing; 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.

If the user's task requires diagnosis first, then entering specialized judgments such as public/organizational/debate/reading, and only finally producing the written piece, first read ../crossframe-suite/SKILL.md for overall orchestration; this skill is only responsible for the article draft and main text generation.

Language principles

Chinese is the authoritative semantics.CrossFrame Essay is only a writing entry point and skill id, and does not carry the authority to explain concepts; English is only used for file names, interfaces, necessary bilingual annotations, or external communication names. When there is a conflict in understanding between Chinese and English, the Chinese terminology, Chinese judgment, and expressions understandable to ordinary Chinese readers shall prevail.

CrossFrame Essay is a parallel writing skill to crossframe , and does not replace crossframe. It converts CrossFrame's structural diagnosis, conceptual fidelity, scale decomposition, and evidence boundaries into critical insight articles aimed at ordinary Chinese readers; when the topic requires deeper expression, it then elevates structural judgments into higher-level concepts, intellectual references, and classical intertextuality. Automatic writing uses the full-visible-v5-longform output tier by default: a fully visible draft + a complete long-form main text. The voice is determined by the crossframe-suite , role, and voice_modepassed in by topic_sensitivity ; when the user explicitly requests a friendly/editorial tone, the modern editorial undertone is enabled, and when the user explicitly requests a neutral report, memo, table, pure diagnosis, or academic abstract, the article voice is disabled.

Core principle: first form a structural insight draft, then write the article main text. Do not skip reasoning and write directly.

Long-form principle: the draft is not a substitute for the main text. After outputting the fully visible draft, the complete article main text must still be written; any task that comes from crossframe-suite and does not explicitly disable the article layer shall be treated as a complete article, and shall not be compressed into a summary, short answer, or bullet-point explanation.

Mandatory sequence

Determine the writing mode:
- Automatic writing: output in one go 结构洞察底稿 and 文章正文, default output_mode=full-visible-v5-longform。
- Interactive refinement: give candidate openings, central propositions, and article skeleton, then advance paragraph by paragraph.
Read ../crossframe/SKILL.md。
Read ../crossframe/references/runtime-read-policy.md and ../crossframe/references/read-routing-map.md, and route the topic to the corresponding CrossFrame protocol.
Read ../crossframe/references/continuity-closure-map.md, at minimum confirm v5-seven-gates-diagnosis-pack and v5-domain-translation-normative-source-pack, and expand their mandatory co-reading closure; for public, intimate, long-term evolution, AI materials, or high-responsibility topics, additionally read the corresponding v5 joint reading package and its closure. When package descriptions are needed, then read ../crossframe/references/continuity-bundles.md or the specific package file in a targeted way.
Use ../crossframe/worksheets/source-continuity-check.md to check whether only isolated concept cards have been read; for in-depth articles, only when source anchors are insufficient, the user requests a source audit, or high-responsibility verification is needed, read the relevant local parts of ../crossframe/references/v5-source-spine.md、../crossframe/references/v5-section-digest-index.md、../crossframe/references/v5-material-selection-map.md or ../crossframe/references/v5-term-fidelity.md in a targeted way.
Reuse the ../crossframe/templates/read-state-capsule.md specified by v5-read-state-capsule; if the upstream has not generated it, return to ../crossframe/SKILL.md to complete it, and do not reinvent source routing within the essay.
Use ../crossframe/worksheets/source-anchor-integrity-check.md to check whether the article's central proposition, mechanism candidates, high-risk concepts, action boundaries, and article translation can point back to the capsule source anchors; content that cannot point back must be marked as "inference in this article / expressive translation / external intellectual mapping."
Read references/evidence-and-search-rules.md and ../crossframe/references/source-ledger-workflow.md, decide whether this session requires internet access or source lookup, and uniformly write it into the source ledger.
Read as needed references/critical-insight-principles.md。
If the topic is an intellectual article, public issue, complex relationship/organizational article, or the user requests depth, conceptual elevation, or classical allusion, read protocols/concept-elevation-protocol.md、references/reference-and-allusion-rules.md and references/concept-reference-map.md。
According to the suite passed in voice_mode Determine whether to read protocols/editorial-comrade-voice-protocol.md and references/editorial-voice-principles.md, and write in the draft 正文声口方案. Only when the user explicitly requests a neutral report, memo, table, pure diagnosis, or academic abstract may the article voice be turned off, and the reason for turning it off must be stated.
During automatic drafting, read protocols/essay-protocol.md; during interactive refinement, read protocols/interactive-drafting-protocol.md。
First generate 结构洞察底稿, write in the draft 文章类型推荐与待选择, but do not first read the writing technique files.
After the draft, confirm the article type: if the user or suite has explicitly specified article_type, record it in the draft and adopt it directly; if it is not specified and the article layer is enabled, you must fully render the nine options of templates/article-type-selection-dialog.md , fill in the recommended option and recommendation reason based on the draft, and wait for the user's reply; if the user replies "default/automatic/whatever," adopt the recommended option in the selector. Do not merely write "article type selector displayed (1-9)."
After the user selects the article type, then read the technique routing table and technique files, and then generate 文章正文: read references/article-technique-routing-map.md, by default read at most 3 core techniques + 2 auxiliary techniques, then read the corresponding references/writing-techniques/*.md files.
Complete the 文章类型与写作技法选择 field in the draft, then translate from the draft into 文章正文。

Reading Rules

By default comply with ../crossframe/references/runtime-read-policy.md: normal drafting does not read evals, examples, complete success/failure cases, the full v5 large index, or the full 50 technique cards.
Automatic drafting: read templates/insight-dossier-template.md and templates/essay-output-template.md; by default execute full-visible-v5-longform。
Interactive refinement: read templates/interactive-session-template.md。
If the topic involves public issues, the latest facts, real organizations, platforms, policies, companies, people, law, technical standards, or data, sources must be checked and a source ledger must be written according to ../crossframe/references/source-ledger-workflow.md ; sources only enter the evidence boundary, counterexamples, real-world cases, and factual limitations, and do not take over the article's thesis.
If the topic is personal relationships, general essays, philosophical concepts, or fictional/generalized material provided by the user, do not go online by default, unless the user requests it or the article needs real sources to avoid misleading.
If conceptual ascent is enabled, first abstract the higher-level concept from the CrossFrame mechanism, then select Chinese and Western classics, historical experience, theory, or literary intertexts, and finally return to real-world judgment.
Automatic drafting first writes 正文声口方案, then drafts. The voice is determined by the voice_mode passed in by the suite:neutral-analysis / neutral-decisive / editorial-reply / editorial-commentary. Only explicit short answers/neutral reports/memos/tables/pure diagnosis/academic abstracts turn off the voice or the long-document position.
First generate 结构洞察底稿, then display the article type selector; the article type selector appears only after the draft and before the main text. The article type only determines main text organization and writing technique reading, and does not change the factual boundary, judgment level, continuous co-reading package, evidence responsibility, or quality gate.
Writing techniques are read on demand only after the user selects the article type. By default, read at most 3 core techniques + 2 auxiliary techniques each time; do not read all 50 technique files. Techniques can only change the expression structure and cannot cross the source anchor boundary of v5-read-state-capsule to add new facts, strong judgments, or the original meaning of the framework.
full-visible-v5-longform By default requires a main text of 1200-2200 Chinese characters, and cannot replace the article opening with "if you only want one sentence," "in plain words," or bullet-point answers.
If the article judgment uses high-risk CrossFrame concepts, read the corresponding concept card according to ../crossframe/references/read-routing-map.md , and use ../crossframe/worksheets/concept-fidelity-check.md to perform a fidelity check.
If the article's judgment triggers the v5.0 continuous section, first follow ../crossframe/references/continuity-closure-map.md to expand the closure, then read the necessary co-reading package files, and write the "source structure continuity check" in the draft.
If the article's central proposition, conceptual ascent, classical intertextuality, or action recommendations cannot refer back to the capsule source anchors, the main text must be written as "this article's inference / expression translation / external thought mapping," and must not claim to be the original meaning of CrossFrame v5. If external sources can only support background or weak signals, the source ledger must also state "what cannot be proven," and must not use the weight of sources to elevate the judgment tier.
If the article uses quotations from classics, conceptual ascent, metaphor, source genealogy, or normative premises, read ../crossframe/references/concept-cards/metaphor-source-transparency.md; direct quotations must be verifiable; when uncertain, only paraphrase or thought mapping may be used.
If the article involves AI compliance, weak signals, inability to exit, absence of institutional infrastructure, instrumentalization, or the exit of open assertions, the corresponding v5.0 co-reading package must be used to complete the reality protection check first, and only then may the article be written.

Hard rules

It is not allowed to write only the main text without producing a draft.
It is not allowed to use retrieved materials to determine the article's stance; retrieval can only corroborate, limit, refute, or add a sense of reality. When real public objects are involved, it is not allowed to write only "sources have been checked"; the source type, the propositions supported, what cannot be proven, the evidence tier, the place of use, and the reason for downgrading must be listed.
It is not allowed to write critique as personal judgment, mockery, moral condemnation, or emotional venting.
It is not allowed to treat terminology as a conclusion. Speak plainly up front, and retain the conceptual chain in the back end.
It is not allowed to fabricate original text, sources, page numbers, or authors' views; when the original sentence is uncertain, only paraphrase or write a thought mapping.
It is not allowed to let classical references take over the article's proposition; quotations can only illuminate real mechanisms and must not overpower the evidence.
It is not allowed to write friendliness as muddled compromise, to write severity as personal judgment, or to use the address "comrade" and slogans to replace analysis.
It is not allowed to write CrossFrame as an all-purpose explanatory machine; when it exceeds its structural judgment capacity, the boundary must be stated.
It is not allowed to write the article as a news roundup, a collage of materials, or an encyclopedia explanation, unless the user explicitly requests this genre.
The paragraph order of the article must follow information dependency: what the reader needs to know first, so that the later judgments can hold.
It is not allowed to treat the complete draft as the main text; after the draft there must be a complete article.
It is not allowed to compress the suite's default article into a short answer of 600 characters or fewer, unless the user explicitly requests a short answer.

Default output

Automatic article writing outputs two continuous parts by default, with the output tier being full-visible-v5-longform / 5.0混合长文：

# 结构洞察底稿

# 文章正文

结构洞察底稿 at least includes:

Analysis object and factual boundaries
Surface phenomena and high-cost signals
CrossFrame routing and this reading
Reading-state capsule summary: source modules, entry continuous co-reading package, must-read-together closure, adjacent candidate packages, downstream reading strategy
Source structure continuity check: triggered continuous co-reading packages, whether the source spine/section-by-section summary was read, whether there is a risk of under-reading
Source anchor integrity check: whether the central proposition, mechanism candidates, high-risk concepts, action boundaries, article-type translation, and writing techniques can refer back to the capsule; how content that cannot refer back is marked or downgraded
v5.0 source structure fidelity and conceptual risk: which concepts cannot be read in isolation, and which adjacent constraints enter this article's judgment
Scale windows and mechanism candidates
Responsibility chain, benefit chain, cost chain
Power, evidence, and weak signal check
Material retrieval and boundaries of evidence
Counterconditions and evidentiary gaps
Conceptual elevation and frame of reference: superordinate concepts, intellectual references, citation methods, sentences that return to reality, citation risks
Main-text voice plan: enable the modern editorial baseline by default; choose reply style/commentary style/neutral expository style; specify the reader's situation, emotional entry point, target of criticism, boundaries of advice, and closing stance
Article type recommendations and items to be chosen: recommended article types, reasons for recommendation, default adopted items
Article type and writing technique selection: after the user chooses, complete the article type, technique files read, main spine, entry technique, structural technique, critical technique, closing technique, and technique execution summary; the summary must record the types of good sentences, relationships between preceding and following paragraphs, micro-uses of the article type, and reverse-checking of failure examples
Source ledger summary: public issues, real institutions, platforms, policies, people, companies, latest facts, and AI/process artifacts must clearly state the source's purpose, evidence tier, what it can support, and what it still cannot prove
The article's central proposition, opening entry point, order of progression, and closing aftertaste

文章正文 At minimum include:

one specific entry point
one clear central proposition
3-5 progressive paragraphs or subsections
add paragraphs on conceptual elevation, classical/theoretical references, and return to reality as needed
switch between reply style or commentary style according to the topic; by default, first catch the question, then give judgment, criticism, and opinions; the explicit neutral expository style may be more restrained, but still must not fall back into piling up concepts
default 1200-2200 Chinese characters; philosophical concepts, intellectual articles, and relational/organizational/public commentary must have buildup, turns, and aftertaste, and must not be written as short answers
at least one boundary, counterexample, withdrawal condition, or evidentiary gap
an ending that does not shout slogans and does not seal the issue shut

Writing temperament

Has sharp judgment, but does not pretend to be omniscient.
Is critical, but preserves evidentiary boundaries and counterconditions.
Can point out the chain of responsibility, but does not compress a complex issue into one person's wrongdoing.
Aimed at ordinary readers; the first paragraph should still be understandable after removing all terminology.
Can respond to readers patiently like a modern editorial comrade: warm but not muddled, decisive but not judgmental.
The ending should have aftertaste; do not replace thought with grand slogans.