Revising a Page

Revision changes a page without producing it again. It starts from the current text, changes only what your instruction and the sources support, and sends the result back through Review. The page keeps its identity and URL, which makes revision the right tool for wording, structure and factual fixes.

Two situations

The page is	How to ask	What happens
A candidate in the current Review	Write a note in the report, or say it in the conversation	The Agent repairs it within the same batch and shows an updated report
Already approved	"Revise <page>: <problem>, keep the source citations"	The Agent runs context revise <page> --instruction "<text>"; the new version enters Review

For an approved page outside the active batch, the revision is queued and runs after the batch by default; --timing priority asks for earlier delivery but never skips Review. Several approved pages can be revised together with context task maintain.

What the Agent does

• Reads the whole page first and keeps explanations that are still correct, including your earlier contributions.
• For a wording-only change, uses the approved text alone; it does not recapture or re-parse the source.
• For a changed fact, reads the relevant registered material. A new fact that only exists in your instruction is first saved as a source, such as a note.
• Edits one section or the whole page. Section edits leave other sections and their citations untouched; a citation change replaces that section's references, which Context re-fingerprints.

Before submitting, the Agent can preview the assembled page and the previous text without accepting the edit.

Moving or renaming a page

context revise <page> --move-to <new-path> revises and moves in one step. The page keeps its article identity and section IDs, its relative links are rebased to the new location, and approval updates incoming Markdown links together — no second copy and no manual deletion. To change only the title or menu position, adjust the knowledge map instead; see Reorganizing at Scale.

Generated reference pages

Pages with generated API tables are regenerated, not hand-edited: --regenerate rebuilds their program blocks from the current source even when the version has not changed. If the same wrong value survives regeneration, the Agent reports the page, field and source comparison instead of repeating the repair.

When the source is wrong

If the page faithfully reflects a source that is itself wrong, correct the source — recapture it, or correct the note or session summary — and then update the page. If the source is right and the page misleads, revise only the page.

修订页面

修订是在不重新生产的前提下修改页面。它从当前文本出发，只改动你的指令和来源所支持的内容，然后把结果重新提交审核。页面会保留其身份和 URL，因此措辞、结构和事实层面的修正都适合用修订来完成。

两种情况

页面状态	如何提出	会发生什么
当前审核中的候选	在报告中写备注，或在对话中直接说明	Agent 在同一批次内修复它，并展示更新后的报告
已批准	“修订 <page>：<problem>，保留来源引用”	Agent 运行 context revise <page> --instruction "<text>"；新版本进入审核

对于不在当前批次中的已批准页面，修订会进入队列，默认在该批次结束后执行；--timing priority 可请求更早交付，但绝不会跳过审核。多个已批准页面可以通过 context task maintain 一起修订。

Agent 会做什么

• 先通读整个页面，保留仍然正确的讲解，包括你此前贡献的内容。
• 仅修改措辞时，只依据已批准的文本；不会重新采集或重新解析来源。
• 修改事实时，读取相关的已登记材料。只存在于你指令中的新事实，会先保存为来源，例如一条笔记。
• 可以编辑单个章节或整个页面。编辑章节时，其他章节及其引用保持不变；修改引用会替换该章节的引用，Context 会重新计算其指纹。

提交之前，Agent 可以预览组装后的页面和之前的文本，而不必接受这次编辑。

移动或重命名页面

context revise <page> --move-to <new-path> 可一步完成修订和移动。页面保留其文章身份和章节 ID，相对链接会基于新位置重新计算，批准时会一并更新指向它的 Markdown 链接——不会产生第二份副本，也无需手动删除。如果只想修改标题或菜单位置，请调整知识地图；参见大规模重组。

生成的参考页面

带有生成式 API 表格的页面需要重新生成，而不是手动编辑：--regenerate 会根据当前来源重建其程序块，即使版本没有变化也是如此。如果重新生成后同样的错误值依然存在，Agent 会报告该页面、字段以及与来源的对比，而不是重复修复。

当来源本身有误时

如果页面忠实反映了一个本身有误的来源，请修正来源——重新采集，或修正笔记或会话摘要——然后再更新页面。如果来源无误而页面有误导，则只修订页面。