@lincyaw/skills 1.0.0-rc.90 → 1.0.0-rc.92

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@lincyaw/skills",
3
- "version": "1.0.0-rc.90",
3
+ "version": "1.0.0-rc.92",
4
4
  "description": "Presmium's bundled authoring, animation, presentation and plugin-development skills for coding agents",
5
5
  "license": "Apache-2.0",
6
6
  "repository": {
@@ -36,7 +36,7 @@
36
36
  ],
37
37
  "dependencies": {
38
38
  "jiti": "^2.7.0",
39
- "@lincyaw/kernel": "1.0.0-rc.90"
39
+ "@lincyaw/kernel": "1.0.0-rc.92"
40
40
  },
41
41
  "presmium": {
42
42
  "skills": [
package/skills/index.json CHANGED
@@ -12,7 +12,7 @@
12
12
  "references/publishing.md",
13
13
  "references/writing.md"
14
14
  ],
15
- "version": "1.0.0-rc.90",
15
+ "version": "1.0.0-rc.92",
16
16
  "hash": "0c47458d29bb2028b093409c03087dc1a20340270525a452d4f771c4257812f9"
17
17
  },
18
18
  {
@@ -546,8 +546,8 @@
546
546
  "references/team.md",
547
547
  "references/timeline.md"
548
548
  ],
549
- "version": "1.0.0-rc.90",
550
- "hash": "3a2f0fb11e1564895c0eef4a7dd9ae4e17d0449202f867f2e22e25fad6009a9b"
549
+ "version": "1.0.0-rc.92",
550
+ "hash": "278048c31db270706e3ac00172af8e181b2425eef4e6d635561873f6edf65e08"
551
551
  },
552
552
  {
553
553
  "name": "presmium-develop",
@@ -557,7 +557,7 @@
557
557
  "references/plugin.md",
558
558
  "references/testing.md"
559
559
  ],
560
- "version": "1.0.0-rc.90",
560
+ "version": "1.0.0-rc.92",
561
561
  "hash": "96422f5cef6a97bba23c9c67b7fb0ee491b18f01f789e262ca24b274e7fabaa1"
562
562
  },
563
563
  {
@@ -567,7 +567,7 @@
567
567
  "SKILL.md",
568
568
  "references/publication.md"
569
569
  ],
570
- "version": "1.0.0-rc.90",
571
- "hash": "dc37b5117120adbc5fc16afc67da8ced0d91992be88751996df133cca596b91c"
570
+ "version": "1.0.0-rc.92",
571
+ "hash": "7a4f66f7ba40f386fd26caaf07c5521394ad579191d212369e567d604938ca55"
572
572
  }
573
573
  ]
@@ -47,7 +47,7 @@ Narration is structured in `narrations`; steps name narration IDs. Cue targets u
47
47
 
48
48
  Validate the edited document (`presmium check` or `check_document` also bundles each scene and reports one that doesn't compile, with its file and line), then look at the actual result. With the MCP server, `capture_slides` gives the rest frame of each step of a page and, with `at`, frames that many milliseconds into a step's animation, drawn by the step clock as a lecture export draws them, so they are reproducible; `measure` returns the boxes of elements (a CSS selector) in slide pixels, for positions and overlaps. Capture the steps in any order you like, since each picture is made from its step and time alone; `arrive: 'backward'` checks that going back to a step shows its frame, and `capture_reading` shows the scene on the reading page, also at a phone width (390). Without the MCP server, build only the relevant form, open it, jump forward and backward in arbitrary order, and inspect the static fallback if it matters (exports print it). Do not claim an animation works merely because JSON validation or compilation succeeded, and do not write browser scripts for pictures the tool takes.
49
49
 
50
- For complete explanations, apply the separate fact, layout, runtime and explanatory-motion checks in [Planning and acceptance](references/acceptance.md#4-review-four-different-things). Review the full timed explanations, beat transitions and arbitrary seeks. A contact sheet verifies stills; it cannot pass motion. The cold viewer must retell which object was operated on, what changed and how the result follows. Static cards, a whole figure with all conclusions permanently visible, or a shared entrance preset used instead of the core operation fail this check even when the data and typography are correct.
50
+ For complete explanations, apply the separate fact, layout, runtime and explanatory-motion checks in [Planning and acceptance](references/acceptance.md#4-review-four-different-things). Review low-cost timed operations and close the independent cold-read findings before freezing the source and resources for the full narrated export. Do not run final export alongside an unfinished review. Retain important captures with `retain: true` and give the reviewer their returned manifest, not a mixed directory of old and new pictures. Timed direct captures use `at`; backward arrival checks resting frames in a separate call without `at`. A contact sheet cannot pass motion: the cold viewer must retell which object was operated on, what changed and how the result follows. Static cards, a whole figure with all conclusions permanently visible, or a shared entrance preset used instead of the core operation fail this check even when the data and typography are correct.
51
51
 
52
52
  ```sh
53
53
  presmium check lesson --json
@@ -34,7 +34,7 @@ Before scene code, save the step sequence and both linked tables below in `story
34
34
  - State what persists across the boundary and where it is at the end and start. When a cut is intentional, explain what the viewer must retain. Replacing the picture with another complete diagram does not establish object continuity.
35
35
  - Use a short “held” entry for intentionally static branding, source text or reading time. A table of final-frame keywords with unspecified actions is incomplete and must be resolved before implementation.
36
36
 
37
- Before polishing, make a rough timed preview of the core operations. The strip of resting frames checks structure; the moving draft checks whether the chosen objects can actually explain the claim.
37
+ Before polishing, make a low-cost timed preview of the core operations: native captures at their planned before/during/after moments, and a short or reduced-resolution clip where continuous motion needs it. The strip of resting frames checks structure; the timed draft checks whether the chosen objects can actually explain the claim. Use this material for the independent review before paying the cost of a full final render.
38
38
 
39
39
  ## 3. Show the operation, not only the final answer
40
40
 
@@ -59,6 +59,12 @@ Entrance effects can support these operations. They cannot substitute for them.
59
59
 
60
60
  Report these results separately in the storyboard and delivery evidence. A pass in the first three does not imply a pass in the fourth. Mark missing observations as unverified rather than turning compilation or a contact sheet into an overall approval.
61
61
 
62
+ Review and production are sequential: close the low-cost timed-preview and independent cold-read findings, then freeze the document, scene code, assets and narration for the full narrated export. Reuse valid speech and generate only what is missing under the existing authorization. Start one full final export from that frozen candidate and inspect its actual media. Do not keep editing or reviewing an unfinished candidate while a final export runs; only an explicitly labelled rough preview may be exported during exploration. A later source or resource change retires the affected review evidence and requires its review again before another final export.
63
+
64
+ Give each reviewer one coherent candidate. For captures needed beyond immediate inspection, use `retain: true` on `capture_slides`, `capture_overview` or `capture_reading`. The response's `capture` identifies the evidence by `id`, `createdAt`, `directory`, `manifestFile`, `retained` and, when available, `source` and `codeBundle`. Read the `capture.json` at `capture.manifestFile`: it records the kind, request, frames, sheets, narration and problems without inline image data. Give the reviewer that manifest and only its listed files. Across captures, match both `capture.source` and `capture.codeBundle`; the same document source alone does not establish the same loaded scene code. If an older host supplies no `capture`, or a needed binding is absent, the candidate match remains unverified.
65
+
66
+ Retention keeps evidence in the service cache's retained area, outside ordinary recent-capture cleanup. It is not project-level long-term archiving or a cloud download link. Do not merge old and new samples in a scratch folder, ask a reviewer to glob it, or write a temporary copy script when native retention is available. A reference or older version used for comparison stays separately labelled. Older evidence never overrides a finding about the current candidate.
67
+
62
68
  | Review | Evidence and question |
63
69
  |---|---|
64
70
  | Facts | Are the values, units, baselines, uncertainty and conditions supported? Are illustrative objects distinguished from measured data? |
@@ -69,9 +75,9 @@ Report these results separately in the storyboard and delivery evidence. A pass
69
75
  For explanatory motion:
70
76
 
71
77
  1. Review each core explanation's full timed preview at the intended pace, including the setup, operation, result and hold. Check its complete time range rather than only an entrance and a middle frame. If continuous playback cannot be observed, report that limitation; timed frame samples and source inspection provide partial evidence, not proof that playback was watched.
72
- 2. Sample the start, the before/during/after states of each planned operation and the settled result using native `capture_slides` with `at`. Choose times from the beat plan, not the same fractions for every shot. Compare against actual narration; a correct final picture does not repair a premature answer or an unexplained transformation.
78
+ 2. Sample the start, the before/during/after states of each planned operation and the settled result using native `capture_slides` with direct arrival and `at`. Choose times from the beat plan, not the same fractions for every shot. Compare against actual narration; a correct final picture does not repair a premature answer or an unexplained transformation.
73
79
  3. Inspect both sides of every core boundary. Check that carried objects are recognizable in position, scale, colour and meaning; identify intentional cuts. Inspect the exported video around those boundaries as well as the preview.
74
- 4. Select steps directly and in a different order, seek into their operations and return backward. Compare the same timestamps. Stable outcomes and intermediate states must come from step and time alone.
80
+ 4. Select nonadjacent steps directly and seek into their operations. Separately capture resting frames with `arrive: 'backward'` and no `at`, then compare them with direct resting frames. `capture_slides` rejects `backward` combined with `at`; these two checks do not establish timed backward playback. Report that only if separately observed. Stable outcomes and intermediate states must come from step and time alone.
75
81
  5. Give an independent cold viewer the timed material and narration, without the source or explanatory notes. Ask: “Which object was being followed? What was done to it? What changed as a result? What comparison or rule supports that result?” Have them retell the chain and predict an appropriate next change. A viewer who repeats the captions or calls it “clear” has not demonstrated this understanding.
76
82
 
77
83
  A contact sheet is useful for facts, layout and the broad story. The still-frame cold read in `team.md` remains an early check; supplement it with the dynamic review above before accepting a complete explanatory animation. Record specific unresolved moments by timestamp and object, the intended operation and the required change. Fix a missing mechanism at the storyboard stage, not by adding effects to its finished card.
@@ -78,9 +78,9 @@ An agent goes through these after every change, on its own pages only:
78
78
 
79
79
  1. **Check the document.** `check_document`, then `get_diagnostics`; fix the problems on these pages. A scene whose code doesn't bundle (a syntax error, a missing module) is reported by `check_document` and `presmium check` as a `presmium.preview.code` error with its file and line; it fails alone, so the other pages still preview and present, and its own page shows the failure (the capture's `problems` say why). Fix it before anything else.
80
80
  2. **Reread the words.** Read your pages' narration, titles and on-screen words in order, as someone who has seen only the pages before them, against the storyboard: is each term's first mention its definition from D, and does no term appear before that step? Does any sentence defend against a reading nobody would make, set up a position nobody holds, or list what the piece does not show (E.14)? Is any word a banned synonym or a calque? Does each step's claim follow from what earlier pages stated, without contradicting them? Rewrite what fails; a sentence kept on purpose goes in the report with its reason. This is a reading, not a search: a marker word found by searching is only a candidate, and the sentence in its place decides.
81
- 3. **Look at every step.** `capture_slides` on each page returns every step's rest frame (12 pictures a call; name `steps` for a longer page). Each must read in 3 seconds, with nothing overlapping or cut off; `measure` with a CSS selector gives the boxes of elements when an overlap is hard to see. Capture again with `arrive: 'backward'` and compare: going back must show the same frame. When the reading page matters, `capture_reading` at 1280 and 390. Change and capture again until every frame passes.
82
- 4. **Check the motion.** Rest frames don't show it. Capture `at` moments inside the steps where something moves, and read the timeline: every move has a reason that means something (a cause, the same object, things converging); delete the moves that have none.
83
- 5. **Report** in this form: for each page, its number of steps and one line describing each rest frame; the picture files; the check result; the sentences kept after rereading the words, with their reasons; problems found in shared modules; where the page departs from the storyboard and why; what isn't done or isn't good enough.
81
+ 3. **Look at every step.** `capture_slides` on each page returns every step's rest frame (12 pictures a call; name `steps` for a longer page). Each must read in 3 seconds, with nothing overlapping or cut off; `measure` with a CSS selector gives the boxes of elements when an overlap is hard to see. Capture again with `arrive: 'backward'` and no `at`, then compare the resting frames. When the reading page matters, `capture_reading` at 1280 and 390. Change and capture again until every frame passes.
82
+ 4. **Check the motion.** Rest frames don't show it. Use direct arrival with `at` for the planned before/during/after moments of an operation, and read the timeline: every move has a reason that means something (a cause, the same object, things converging); delete the moves that have none. Do not combine `backward` and `at`, or describe separate backward rest and direct timed captures as a timed backward test.
83
+ 5. **Report** in this form: for each page, its number of steps and one line describing each rest frame; the retained capture manifest and files for this candidate; the check result; the sentences kept after rereading the words, with their reasons; problems found in shared modules; where the page departs from the storyboard and why; what isn't done or isn't good enough.
84
84
 
85
85
  ## 5. The coordinator during production
86
86
 
@@ -88,7 +88,7 @@ An agent goes through these after every change, on its own pages only:
88
88
  - Read every report and look at its pictures. Settle a difference between pages (one part drawn two ways, one thing named two ways) at once: decide, write it into the storyboard, tell the agents concerned.
89
89
  - Forward a pitfall one agent hit to every agent still working, at once; finding it nine times costs more.
90
90
  - After each batch of reports, read the narration of the whole document in order for terms that one page uses before another page defines them: such a term belongs to neither agent alone. Settle it in the term table and tell both.
91
- - When every page is done, run the cold read (section 7) before the author sees the piece.
91
+ - When every page is done, run the independent cold read and low-cost timed-operation review (section 7 and `acceptance.md`). Close their findings, freeze the document and referenced resources, then run the full narrated export and inspect its actual media. Review and final export do not run in parallel; an explicitly labelled rough preview during exploration is the exception. Automatic workflows perform these reviews internally without adding a routine user approval gate.
92
92
 
93
93
  ## 6. Review rounds
94
94
 
@@ -115,21 +115,21 @@ A second explainer (24 pages, eight agents) added three rules in the middle of p
115
115
 
116
116
  A fresh agent that took no part sees only the pictures and the narration, in order, and reports what a first-time viewer gets. It finds what the makers can't see: contradictions between pages, one word with two meanings, names that drift, concepts used before they were introduced or never defined, sentences that carry nothing, elements nobody explained. In the explainer it found three problems that blocked the main line, more than twenty concepts left unclear and a dozen flaws in pictures, most of which the agents' own page-by-page checks had missed.
117
117
 
118
- **When**: after every round of production or fixes, before the author reviews. A new reviewer each round: one that saw the last version already knows the answers.
118
+ **When**: after a round of production or fixes, before freezing the candidate for a full final export or showing the requested preview. Review low-cost material first; do not start the final render while these findings are still open. A new reviewer each round: one that saw the last version already knows the answers.
119
119
 
120
120
  **Prepare the material:**
121
121
 
122
- 1. `capture_overview` with `steps: 'all'`, `captions: false` and `images: false` takes every step of every page at rest, the whole slide without the captions over it, into one directory of the service's state, named so that sorted names are the playing order: `03-12-game.png` is page 3, step 12 (pages from 1, steps from 0, as the tools count them). Copy the directory at once: the service keeps only its latest captures. Without `steps: 'all'` it takes each page's last step, and its sheets of thumbnails show the whole piece at a glance.
123
- 2. A frame taken mid-motion gets reported as a flaw: these are rest frames, drawn from the step alone. Numbers in names must be zero-padded (the tool does it): with `15-1.png` and `15-10.png`, step 10 sorts right after step 1.
124
- 3. The same directory has `narration.md`, each picture's narration under its file name: the reader looks at whole slides and reads the script as text, instead of small captions over the picture.
125
- 4. Give nothing else: no source, storyboard, term table or notes, or the reader is no longer cold.
122
+ 1. `capture_overview` with `steps: 'all'`, `captions: false`, `images: false` and `retain: true` takes every step at rest and retains this review's evidence in the service cache. Without `steps: 'all'` it takes each page's last step. Read `capture.manifestFile` and use its exact file list; do not copy captures into a shared scratch directory or scan unrelated PNGs already there. Keep samples from a different document, scene-code or resource version out of this candidate's review set.
123
+ 2. Label rest frames and timed samples separately. A deliberate intermediate state must not be judged as a failed rest frame. For motion review, retain direct `at` samples around each operation; for reverse navigation, make separate `arrive: 'backward'` rest-frame captures without `at`.
124
+ 3. Supply the corresponding narration and the manifest identifying this candidate's source, request context and files. The overview's `narration.md` associates each rest picture with its words. Match both `capture.source` and `capture.codeBundle` across the supplied captures, as described in `acceptance.md`; source alone does not verify scene-code identity. A missing `capture` or binding leaves that match unverified: obtain coherent evidence before requesting a current-version verdict.
125
+ 4. Give no source code, storyboard, term table or explanatory notes to the cold viewer. The manifest identifies the evidence; it does not supply the intended interpretation. Supplement the still-frame passes below with the timed material and dynamic questions in `acceptance.md` before accepting explanatory motion.
126
126
 
127
127
  **First pass: understanding.** Replace the angle brackets; write the prompt in the document's language when the report should be in it.
128
128
 
129
129
  ```
130
130
  You are seeing this explainer for the first time, as a viewer and a reviewer. You are <the audience, e.g. a researcher who evaluates LLM agents and knows incident troubleshooting, but hasn't seen this design>. You took no part in making it. Don't read any source code, storyboard or notes; use only the material below.
131
131
 
132
- Material: <N> PNG pictures in <folder>, the rest frame of each step of the presentation. File names are <page number>-<step number>-<page key>.png (steps from 0), and their sorted order is the playing order. <folder>/narration.md has each picture's narration under its file name: read it with the picture, as a viewer hears it. Look at every picture in order with the Read tool; skip none.
132
+ Material: <retained capture manifest and its listed rest-frame PNGs>, the rest frame of each step of one candidate. Read only those listed files, in their recorded presentation order, and the corresponding narration. Do not glob the directory or mix other captures into the verdict. Timed samples are separately identified and are not rest frames. Look at every listed rest picture in order with the Read tool; skip none.
133
133
 
134
134
  First, page by page. After each page (all pictures with the same page number), write:
135
135
  - what the page says, in one sentence of your own;
@@ -174,4 +174,4 @@ Add these to the problem list; a contradiction in the argument is a P0.
174
174
  2. A problem with the method itself, such as a rule whose reason no longer holds, goes to the author; don't decide it.
175
175
  3. The wording problems become one term table in the storyboard, which every page follows; definitions and calques go into its columns, which the next round's agents and reviewer read against.
176
176
  4. Fixes go to the agents that made the pages.
177
- 5. After the fixes, capture everything again and give it to a new reviewer.
177
+ 5. After the fixes, make a retained capture set for the new candidate and give its manifest and material to a new reviewer. Retire the affected old evidence; do not overwrite a current conclusion with pictures from the prior round.
@@ -36,9 +36,9 @@ Presmium 原生 MCP 是创建、编辑、检查、配音、导出和截图的首
36
36
  - 有指定参考作品时,先拆解其实际动作、空间和对象连续性,再设计新片。在本仓库的 `examples/papers/` 系列,第一期 `papers/hinted-scaling-laws/` 是逐步解析方式的参考:读其场景时间线并检查动态画面,不只复用品牌。其他工作区采用其指定参考,不把这一篇当作所有动画的唯一风格。
37
37
  - 每个核心解释都展示设置、操作、结果与停留,按内容安排语义节拍;概念由具体对象和可见操作引入再命名,结果交代基线、指标和关键条件。静态卡片、原图加全部结论常亮、通用入场统一套用不能替代解析演示。
38
38
  - 优先用论文原图做镜头、标注和对照,动画表达机制或因果,不堆说明卡片。无法复算的数字引用论文明确位置;示意与估算在教学画面中标明。
39
- - 通过 MCP 保存当前文稿和场景,检查并看静态联系表及过渡帧。连续场景的主体保持存在,新增信息才入场;步骤能任意跳转和回退。
39
+ - 通过 MCP 保存当前文稿和场景,先用低成本的定时关键操作预演和联系表审阅;连续场景的主体保持存在,新增信息才入场。重要 capture 用 `retain: true` 留在服务缓存,引用 `capture.manifestFile` 中本轮实际文件;按链接规范同时核对 `capture.source` 与 `capture.codeBundle`,只交给审阅者同一候选版本的证据,不混合旧图或编写临时复制脚本。旧主机缺少这些依据时记为来源未验证。定时采样用 direct 加 `at`,回退休止帧用 `arrive: 'backward'` 且不带 `at`,两者分开验证。
40
40
  - 片头从完整系列封面开始,开场配音保留;用页面级 `captionVisible: false` 隐藏片头字幕,正文恢复字幕。不删讲稿或把字幕隐藏当成静音。
41
- - 预演自检通过后继续制作。协调者统一编辑文稿,场景文件明确分工。联系表冷读检查事实、词语与布局;独立的动态审阅另看完整核心段落、操作边界与任意 seek,要求冷读者复述对象、操作及结果的关系。四类验收分别记录,不能用静态审阅或技术通过代替解析动画质量通过。
41
+ - 协调者统一编辑文稿,场景文件明确分工。先完成独立冷读和动态预演审阅,闭环问题后冻结文稿、场景代码、素材与讲稿,再进入正式配音导出。联系表检查事实、词语与布局;动态审阅要求冷读者复述对象、操作及结果的关系。四类验收分别记录,不能用静态或技术通过代替动画质量通过;没有实际观察的倒退定时播放也不记为已验证。
42
42
 
43
43
  ## 3. 配音与真实时长
44
44
 
@@ -50,7 +50,9 @@ Presmium 原生 MCP 是创建、编辑、检查、配音、导出和截图的首
50
50
 
51
51
  ## 4. 导出并验收成片
52
52
 
53
- 用当前文稿的原生视频导出动作,烧入中文字幕并导出联系表。检查实际视频文件、声音、首帧、正文字幕、关键结果帧、相邻换页和结尾,并按[动态验收](../presmium-animate/references/acceptance.md#4-review-four-different-things)核对分镜中的完整操作链、结果时机与对象连续性;事实、布局、运行正确不等于动画质量通过。源文稿、预计输出路径或静态联系表不算成片验收。成片须有完整配音;无声预演不是这条流程的最终交付。
53
+ 从问题已闭环的冻结版本启动一次完整配音成片导出,烧入中文字幕并导出联系表。审阅与正式导出串行;只有明确标记的低成本粗预演可以在探索阶段导出,不边修改边跑完整成片。导出期间来源变更时,先重新核对受影响的预演与配音绑定,再决定下一次正式导出。
54
+
55
+ 随后检查实际视频文件、声音、首帧、正文字幕、关键结果帧、相邻换页和结尾,并按[动态验收](../presmium-animate/references/acceptance.md#4-review-four-different-things)核对分镜中的操作链、结果时机与对象连续性。事实、布局、运行正确不等于动画质量通过;源文稿、预计路径或静态联系表不算成片验收。成片须有完整配音,无声预演不是最终交付。
54
56
 
55
57
  ## 5. 独立封面与简介
56
58
 
@@ -56,12 +56,12 @@
56
56
  ## 浏览器操作
57
57
 
58
58
  1. 读当前浏览器控制工具的说明,用用户指定浏览器或现有已连接浏览器。只操作本次所需页面。进入小红书创作服务平台,核对登录账号和原笔记/候选期数。
59
- 2. 新笔记上传最终 MP4;修改已存在的同一篇时优先编辑原笔记,不重新发一条。上传先按工具的文件上传能力操作。若扩展文件上传要求新增“文件网址”访问权限,优先使用系统原生文件选择器,不擅自扩大扩展权限。路径用实际绝对路径;等待原生“前往文件夹”输入框出现后再填路径。
59
+ 2. 新笔记上传最终 MP4;修改已存在的同一篇时优先编辑原笔记,不重新发一条。替换视频用原笔记的“重新上传”,先等现有封面、合集等字段载入,再改动。上传先按工具的文件上传能力操作;隐藏的文件输入框无响应时,改用页面可见的上传入口。若扩展上传要求新增“文件网址”访问权限,优先使用系统原生文件选择器,不擅自扩大扩展权限。原生操作前先核对前台是目标标签页;路径用实际绝对路径,等待“前往文件夹”输入框出现后再填入。
60
60
  3. 在“编辑封面/设置封面”的自定义图片入口上传 `cover.png`。完整检查当前比例、缩放和裁切后的实际图片:品牌、主题、图和边缘文字均可见。平台默认第一帧、智能推荐或付费 AI 封面都不替代已制作的独立封面。
61
61
  4. 填最终标题与简介,核对可见输入值、长度和预览。编辑器可能是 contenteditable;使用真实可编辑字段而非带占位文本的空壳。话题按当前 UI 处理,不使用过期元素序号。完整素材链接、来源说明和发布记录不进入公开简介。
62
- 5. 合成配音按平台实际入口设置“笔记含AI合成内容”等内容类型声明。它是平台字段,和简介中的制作注释是两回事。公开可见、即时发布;其他合集、地点、群聊、原创声明、活动或付费推广只有在相应授权下设置。
62
+ 5. 工作区已指定或用户已选择系列合集时,核对并复用已有合集;新笔记加入它,替换视频后再核对归属,不另建同名合集。合集名称保存在工作区约定和发布记录中,不把某个账号的名称写成技能全局默认。合成配音按平台实际入口设置“笔记含AI合成内容”等内容类型声明。它是平台字段,和简介中的制作注释是两回事。公开可见、即时发布;其他合集、地点、群聊、原创声明、活动或付费推广只有在相应授权下设置。
63
63
  6. 具备发布授权后提交;若页面提出验证码、法律协议或新的权限要求,按浏览器工具的确认规则在该具体动作处处理。缺少登录或需用户完成验证时保留页面并明确接手点。不要用脚本绕过。
64
- 7. 核实成功回执与笔记管理中的那一条记录,保存可显示标题、封面和状态的截图。提交成功且审核中即完成本次上传工作,最终状态照实写入记录。交付笔记链接和实际素材。
64
+ 7. 核实成功回执与笔记管理中的那一条记录,保存可显示标题、封面和状态的截图。替换视频时同时记录提交前显示的新文件名、实际素材哈希及原笔记ID,避免把“打开编辑页”记作替换成功。“已发布”列表可能暂时不显示更新后正在审核的笔记,切到“审核中”或“全部”核实,不据此重新发一条。提交成功且审核中即完成本次上传工作,最终状态照实写入记录。交付笔记链接和实际素材。
65
65
 
66
66
  每次交互后读取最新状态,只有需要判断裁切或布局时补截图。浏览器自动跳转或新开标签时先查当前标签列表再选择,不因原标签没变化而重复点击。遵守当前控制工具要求的完成摘要与标签保留流程。
67
67