Skip to content

armgit / hypit

Public Imported from GitHub
0

Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views.

docs: place check sampling rules with route checks (#127)

rponeawa6e302f197049
diff --git a/docs/guide/skill.md b/docs/guide/skill.md
index bfb2bf8d..30362310 100644
--- a/docs/guide/skill.md
+++ b/docs/guide/skill.md
@@ -15,30 +15,6 @@ The important distinction is:
- tools decide whether the written project is legal, wired, covered, measurable and recoverable;
- a visual observer reports differences, but the agent decides whether a report is a real defect.
-## Two extraction rules that keep the checks complementary
-
-This is a central design rule for both `original-authoring` and `reconstruction`:
-
-| Check | What it extracts | What it answers | Why this rule |
-|---|---|---|---|
-| Visual review | One **full clip** at the first appearance of each distinct visual declaration/style change | Does the complete look and motion match the brief (original) or the reference (reconstruction)? | A full clip preserves entrance, movement, exit, transient overlap and appearance changes. Rechecking every caption text or every timestamp would add repetition without adding a new visual declaration. |
-| Mechanical layout | One frame from the **longest stable interval of each `Present`** | In a settled, actually rendered DOM, do content and containers fit and align? | Sampling during motion would report normal animation as overflow or offset. Sampling per `Present` means separate caption Cues/content states are measured independently. |
-
-These rules are deliberately different and must not be swapped. The visual plan does not use the
-longest-stable frame; it reviews the whole clip. `layout_check` does not judge visual similarity; it
-measures the realized composition and returns candidates for the Agent to interpret. Content or time
-extremes are not additional visual-selection rules: a stable overflow is found by the per-`Present`
-layout measurement, while a transient or media-internal problem remains the responsibility of the
-full-clip visual review.
-
-The extraction is shared across the two creation routes. Only the judging basis changes:
-
-```text
-original-authoring: visual clip → compare with brief
-reconstruction: visual clip → compare with reference
-both routes: Present stable frame → measure DOM → Agent judges the candidate
-```
-
## How a request becomes a video
```text
@@ -175,13 +151,23 @@ records an intentional geometry decision with `layout_accept`, then reruns `layo
candidates are resolved or accepted. A relevant Source, Recipe, package, font or runtime change
invalidates affected evidence.
+The two extraction rules used by both creation routes are:
+
+| Check | Extraction rule | Why |
+|---|---|---|
+| Visual review | For each distinct visual declaration/style change, take the **full clip at its first appearance**. | The full clip preserves entrance, movement, exit and transient visual differences. It avoids rechecking every caption text or timestamp when the visual declaration is unchanged. |
+| Mechanical layout | For each `Present`, take one frame from its **longest stable interval**. | A settled frame measures actual DOM geometry without mistaking normal animation for overflow or offset. |
+
+Do not swap these rules: visual review judges the complete clip against the brief or reference, while
+`layout_check` only measures realized layout and returns candidates for the Agent. Content/time extremes
+are not additional selection rules; stable overflow is found in the per-`Present` measurement, and
+transient or media-internal problems remain visible to the full-clip review.
+
### 6. Creation-route visual review and repair
Original authoring calls `authoring_check`; reconstruction calls `reconstruction_check`. Each command
-reads Source and creates a deterministic visual-review plan. It selects one full clip at the first
-appearance of each distinct visual declaration (including each style/declaration change), rather than
-asking an agent to invent an arbitrary render list. The longest-stable-interval sample belongs only to
-the separate hidden-browser `layout_check`, not to this visual review.
+reads Source and creates the visual-review plan described above. The longest-stable-interval sample
+belongs only to the separate hidden-browser `layout_check`, not to this visual review.
For each plan entry:
diff --git a/docs/zh/guide/skill.md b/docs/zh/guide/skill.md
index 9436485f..b641d151 100644
--- a/docs/zh/guide/skill.md
+++ b/docs/zh/guide/skill.md
@@ -14,28 +14,6 @@ description: Hypit skill 如何把描述或参考视频变成可检查、可恢
- 工具决定项目是否合法、graph 是否接通、布局和证据是否稳定;
- 视觉 observer 只报告差异,是否算问题由 agent 判断。
-## 两条必须先看懂的抽取规则
-
-这是 `original-authoring`(原创制作)和 `reconstruction`(复刻)共用的核心设计:
-
-| 检查 | 抽取什么 | 回答什么问题 | 为什么这样做 |
-|---|---|---|---|
-| 视觉检查 | 每种不同视觉声明/样式变化在首次出现处的一个**完整片段** | 画面和运动是否符合 brief(原创)或参考视频(复刻)? | 完整片段能保留出现、移动、离开、瞬时重叠和外观变化。无需因为每句 caption 内容或每个时间点不同而重复检查。 |
-| 机械布局检查 | 每个 `Present` 的**最长稳定区间中的一帧** | 在真实渲染的稳定 DOM 中,内容和容器是否越界、偏移或重叠? | 动画过程中的偏移是正常运动,不应被误报为布局问题;按 `Present` 分开取样可以分别测量不同 caption Cue/内容状态。 |
-
-这两条规则职责不同,不能互换。视觉计划不使用最长稳定帧,而是检查完整片段;`layout_check`
-不判断视觉相似度,只测量真实 composition 并把结果作为候选交给 Agent。内容极值和时间极值不再
-是额外的视觉选片规则:稳定状态下的越界由逐个 `Present` 的布局测量发现,瞬时运动问题或媒体
-内部的问题仍由完整片段的视觉检查负责。
-
-两条创作路径共用同一套抽取方式,只有判断依据不同:
-
-```text
-原创制作:视觉片段 → 对照 brief
-复刻: 视觉片段 → 对照参考视频
-两者: Present 稳定帧 → 测量 DOM → Agent 判断候选是否真是问题
-```
-
## 用户请求如何变成完整视频
```text
@@ -156,9 +134,18 @@ hypit-reference-video-tools layout_check --run <run>
布局发现只是交给 agent 的辅助证据,不是自动真相。agent 修复真实问题,或用 `layout_accept` 记录有意几何,然后重跑 `layout_check`,直到候选消失或被接受。相关 Source、Recipe、包、字体或 runtime 改动会使证据失效。
+视觉检查和机械布局检查使用两条不同的抽取规则:
+
+| 检查 | 抽取规则 | 为什么这样做 |
+|---|---|---|
+| 视觉检查 | 每种不同视觉声明/样式变化,在首次出现处取一个**完整片段**。 | 完整片段保留出现、移动、离开和瞬时视觉差异;样式不变时无需因为每句 caption 或每个时间点重复检查。 |
+| 机械布局检查 | 每个 `Present` 在自己的**最长稳定区间**取一帧。 | 稳定帧才能测量真实 DOM,避免把正常动画过程误报成越界或偏移。 |
+
+两条规则不能互换:视觉检查对照 brief 或参考视频判断完整片段;`layout_check` 只测量真实布局并把结果作为候选交给 Agent。内容极值和时间极值不是额外选片规则:稳定越界由逐个 `Present` 的测量发现,瞬时或媒体内部的问题由完整片段的视觉检查负责。
+
### 6. 创作路径的视觉检查与修复
-原创制作使用 `authoring_check`,复刻使用 `reconstruction_check`。命令读取 Source,按每个不同视觉声明(包括样式/声明变化)的首次出现生成一个完整片段的视觉检查计划,不让 agent 随意编一份 render 列表。最长稳定区间取样只属于另外的隐藏浏览器 `layout_check`,不属于视觉检查。
+原创制作使用 `authoring_check`,复刻使用 `reconstruction_check`。命令读取 Source,按上面的视觉规则生成检查计划。最长稳定区间取样只属于另外的隐藏浏览器 `layout_check`,不属于视觉检查。
每个计划项按以下闭环执行: