aikey-cut

This skill was originally named /zmm-…, and from v1.0.0 it is renamed to AI KEY（/aikey-…); the old name is no longer updated. Existing users upgrading see README。

aikey-cut: Talking-head video editing

It handles "after shooting." Before shooting, topic selection, script writing, and review belong to /aikey-topic /aikey-script /aikey-review。

🔴 Boundary: only delete and reorder, never add words (clarified 2026-09-05)

Here we only handlethe version he actually said. Within that version:

✅ Allowed ❌ Not allowed

Delete (dead shots, repeats, filler words, jargon, trailing fragments) Addany word he did not say

Reorder(adjust order at the container level, bring the conclusion to the front) Smooth sentences, change word order, replace wording

Merge pickup shots into the original footage Reorder the segment where he derives the conclusion (see references/内容层重组.md §II.3)

⚠️ "This skill does not change the script" means no words are added, not that nothing may be touched. Deleting and reordering according to the script rulesis exactly this skill's main battleground。

✅ Allowed	❌ Not allowed
Delete (dead shots, repeats, filler words, jargon, trailing fragments)	Addany word he did not say
Reorder(adjust order at the container level, bring the conclusion to the front)	Smooth sentences, change word order, replace wording
Merge pickup shots into the original footage	Reorder the segment where he derives the conclusion (see `references/内容层重组.md` §II.3)

0. Startup order (do not skip)

🔴 Probe the environment first, then decide how to proceed(corrected 2026-09-05):
- Host with the ChatCut plugin prerequisite skill → load it first,do not bundle it in the same turn as ChatCut tool calls
- Desktop MCP direct connection(tool names look like chatcut_desktop/*）→ Without that prerequisite skill, per server requirements first run a read-only locate (get_active_project) to confirm the target project
- None of them / reports signed out → follow references/ChatCut实操.md thefallback pathin §I: notify to download + local transcription bypass(none of the content-level work requires ChatCut)
chatcut-talking-head-guide — execution rules for talking-head editing
Read aikey/references/家族公约.md(if unreadable → state explicitly "the convention cannot be read, red lines cannot be guaranteed" and stop)
- aikey/references/交互规范.md（🔴 not just read once: at wrap-up follow the §IV trio — Recap · Before/After · next step as numbered options; for missing info follow §IV withmultiple-choicequestions,ask only one at a time; for cases where it does not apply see §V)
Read {vault}/00-规则与索引/口播拍摄基线.md(time/place/outfit fixed once and for all)
Read memory {vault}/08-技能记忆/aikey-cut/ + _通用/
🔴 When content must be changed (not just removing verbal tics), read references/内容层重组.md
🔴 Before calling ChatCut tools for the first time, read references/ChatCut实操.md(three tool pitfalls found in real testing)
- Whengenerating B-roll / effects / transitions, then read references/生成素材.md
🔴 After cutting, you must return to 口播稿输出格式.md the "upload version" under §四 — see §五 of this document

This skill's built-in criteria are in references/规则卡.md(criteria / why / how to check / strength); read them once before starting;{vault} when there is a corresponding rules file in the vault, the vault prevails and the rule card serves as the base.

⚠️ Read on demand; do not read everything at once. This skill covers only the "filmed → final cut" stage; you do not need to read files from the topic-selection layer or the scriptwriting layer.

1. Account default parameters (🔴 fixed; do not ask the user again)

🟢 Godfather of Australian AI, set on 2026-08-27.These are default values, not questions that need to be confirmed every time. If the user has not said to change them, use them directly; if the user says something else, what the user says this time prevails.

⚠️ The table below contains measured values for "this one account", not general recommendations. If there is config , thenread from config Read; if it cannot be read, thenfirst ask whether it is the 澳洲AI教父 account: if yes, use the table below; if not, ask item by item and do not apply it. When changing accounts / changing topics, follow the main workflow as usual,only swap this table。

Item	Default value
Platform	Douyin (vertical 9:16)
Final cut duration	Keep the original length — do not cut content just to make it short (the criterion is absolute watch time, not completion rate)
Overall speed	🔴 1.15×(uniform speed-up after speech editingafter, not adjusted segment by segment)
Default processing	① Speech editing(remove filler words, pauses, repetitions) ② Subtitles
Not done by default	B-roll · MG animation · background music — only do them if the user explicitly asks
Pacing style	Tight and punchy

Subtitle specifications (🔴 fixed)

Item	Value
Font	HarmonyOS Sans Bold
Outline	Slight
Shadow	Yes, light
Overall look	Clean and modern
Highlight	★ Yes, color yellow `#FFE20A`
🔴 Who selects the highlighted keywords	The user selects them. When producing the cut, provide candidates;do not decide for them
🔴 Highlight method	Highlighted-word mode: only the selected words / phrases turn yellow; the rest stay white. 🔴 Do not use the read-progress style(where everything before the current point turns yellow, and the whole sentence turns yellow at the end). Among ChatCut's `highlightUnit` three values, `word` / `line` are all progress-style, so set `off`, then color the selected cards individually

2. Workflow (seven steps; you must stop between each step)

🔴 talking-head-guide Hard rules:There are dependencies between multiple processing steps; the final version must be set in order, and after each major step you must confirm with the user separately; do not pack multiple checkpoints into one reply. If the upstream changes, everything downstream has to be redone (subtitles written against the pre-speedup timeline = completely wasted).

⓪ 素材清点与合并 → 停，说清有几条素材、哪条是补拍、要不要合并
① 素材进项目     → 停，确认转录出来了
② 语音剪辑       → 停，确认「他实际说的」这一版对不对
③ 整体加速       → 停，确认节奏（倍速见 §一）
④ 字幕 + 高亮     → 停，让用户选高亮关键词
④b B-roll / 生成素材（用户明确要才做）→ 停，确认配哪几句、用真素材还是生成
⑤ 交付           → 只有用户明确说要导出/下载才导
⑥ 导出后验文件本身 → 抽帧确认字幕/B-roll/打码都烧进去了

⓪ Stocktaking and merging of materials (added 2026-09-05)

The skill originally assumed only one piece of material by default. In practice there are often two or more(original footage + pickups, multi-camera, multiple recordings).

Firsttranscribe each one, read them separately,do not splice right away
State clearlywhich one is the main material and which one is the pickup; the pickup mustcover which segment
🔴 Once pickups are brought in, you must do a 'verbatim repetition scan' — see references/内容层重组.md §II.2
Mergearrange in the new order, not by appending the pickups to the end

① Materials go into the project

First list_projects / create_project / target_project lock down the project,then do anything else
As soon as the project is createdimmediatelybring up the editor (preview_start), so the user can watch the progress
Materials can come in three ways: the user uploads directly in the editor · a local path is given in the session (go through asset-import) · the user has already uploaded it (first use browse_assets to look,don't just say you can't find it right away）
Once the transcription is out you can start work, no need to wait for the raw bytes to finish transferring (A-roll only needs the text transcript)

② Speech editing (the main battleground of this skill)

🔴 This step has two layers; do not mix them together: Mechanical layer(fixed verbal tics, batch pauses) does not require understanding the meaning; a tool can do it; Content layer(what to keep, what to cut, what to move)requires understanding the meaning, the criteria are in references/内容层重组.md。

⚠️ Doing only the mechanical layer = just wiping the film clean, not improving it. When the user says 'cut it', they usually mean both layers,but every cut or change in the content layer must have its reason stated clearly。

Strictly follow talking-head-guide 's A-roll process:

read_script → read through timeline.mdonce to see the structure clearly
First run the mechanical cleanup（clean_script): only handle fixed verbal tics (uh/um/mm) and long pauses
🔴 After it runs, you must re-read the refreshed timeline.md — what you read before is now out of date
Then do thecontent layerjudgment — follow references/内容层重组.md 's five criteria: identify dead footage · verbatim repetition scan(mandatory for pickups) · do not reorder the reasoning chain for suspense · the opening must have both a hook and resonance · before cutting, ask 'will it still be clear after cutting this?'
apply_script land it on the timeline
Read it back and check:seam logic(is a bridge missing) · consistency before and after · does the relative time still hold · is the ending complete · are the pauses too tight

🔴 Three things specific to this account; don't cut them out by general rules

Keep	Why
Spoken-language noise(呢 / 啊 / 哈 / reduplication / slip-of-the-tongue repetition)	Rule Every sentence being perfectly neat = the audience can tell in one second that it's being read from a script.⚠️ Generic "de-tic" cleanup will wipe these out along with everything else —— keeping one every 5–8 lines is deliberate, not a mistake
Inserted lines that invite self-reflection("Think about it carefully" "I wonder if you've ever had a moment like this")	Rule It carries an action instruction, not a filler word
Qualifiers(mostly / a lot of the time / it may well be)	Rule Must not be changed into absolutes

🔴 Criterion: After deleting this word,does the sentence become cleaner, or does it start to sound like a script? Sounds like a script = don't delete.

Pause guidelines

Clearly long pauses (>0.8–1s) → compress to about 0.3s
Keep 0.3–0.5s between sentences —— don't compress them away
Keep pauses at turns, contrasts, and emphasis; those are deliberate (the spots marked ⏸⏸ in the script)
Normal breathing within a sentence → leave untouched

③ Overa11 speed-up 1.15×

🔴 Must be done after the voice edit is finalized, because if the order is reversed, all subtitles will have to be redone. Apply uniform speed-up,not segment-by-segment speed adjustment —— adjusting per segment makes the intonation speed up and slow down unpredictably.

④ Subtitles + highlighting

Follow the captions spec of talking-head-guide , and apply the fixed spec in §I.

🔴 I must not decide the highlighted keywords for him. Method:

Read through the subtitles and pick outone every 8–12 secondscandidate keywords (pick the core words in judgment sentences, numbers, transition words, and punchline sentences)
List them for the user to choose, clearly marking at which second and what the original sentence is
Apply highlighting only after the user has checked them off

⚠️ Do not highlight whole sentences —— the purpose of highlighting is to make the eyes pause while scrolling; highlighting a whole sentence is the same as no highlighting.

④b B-roll / generated assets (only do this if the user explicitly asks)

Place after speed-up and subtitles: only once the timeline is set do you know how many seconds each sentence takes.

Choose one of two asset sources: the user's own real footage (filter per references/ChatCut实操.md §IV) / generated
To generate → per references/生成素材.md：
- judge sentence by sentence whether generated footage fits.🔴 Never pair generated footage with testimony-type content: "I've done this" "this really happened," and data screenshots all count
- Give the user a list to check off, then write the prompts after they've checked
- Hand off to chatcut-video-gen to produce the footage (family convention: "centralized invocation").🔴 Before submitting, report the number of clips, duration, and model, and wait for him to confirm the cost
After the footage is produced, review each clip (garbled text, distorted hands and faces, whether the texture matches), and don't force-use subpar ones; after placing them on the timeline,mute the generated footage(Seedance always comes with sound), and trim the cut length per references/ChatCut实操.md §IV
🔴 Publishing reminder: Turn on the platform's "AI-generated" declaration; do not remove the generation tool's watermark

⑤ Delivery

🔴 By default, what is delivered is an "editable ChatCut timeline", not an MP4. Only when the user explicitly saysexport / render / download / final deliverydoes it go through submit_export → track_export。 "Trim it a bit", "clean it up", and "make a version" do not count as intent to export.

2b. 🔴 Acceptance discipline: metadata lies (added 2026-09-05, after three real tests)

Row counts, text, and visuals can only be judged by full-size rendered frames.

What fooled us	Actual
The subtitle API reported "one line, no overflow"	It wrapped to two lines when rendered
In the multi-grid thumbnail, the text looked wrong	A full-size single frame proved it was a misread
The tool receipt reported "cleaned up 1 word"	The timeline was actually 4.9 seconds shorter

After every major step, always verify two things：

Numbers — do the frame count / duration / segment count match expectations
Visuals — renderfull-sizeframes,don't draw conclusions from thumbnails

⚠️ Thumbnails are only for coarse screening of "which frame to pick". See references/ChatCut实操.md §3.

3. Red lines (unique to the editing stage)

	Rule
Don't change the content	The script was finalized at shoot time. Editing only cuts, never adds;don't fill in words for him, smooth sentences, or change word order
Visual information points	Outfits / desk setup / background / wall decor are all information points.When editing, don't cut out shots that carry information
No AI screenshots	For the segment about AI, the visualsshould not use AI interface screenshots — that narrows the audience (`不要变成工具号`）。same for generated assets: don't generate visuals of "AI at work"
Generated assets don't serve as proof	AI-generated visuals may only be used for analogy, process, atmosphere, and transitions,and cannot serve as proof of "I did this / this really happened" or any data; turn on the AI-generated declaration when publishing
Where to place large text	In the script's "points that need handling", lines marked ⏸ for large textmust match up during editing
Mask amounts	Any visuals showing backend data / payments / client information must be masked;specific income amounts follow the same standard in visuals, spoken lines, subtitles, and titles, and appear nowhere(see `references/规则卡.md`"Visual masking checklist")

4. Who it's for

One person shoots and edits alone. He does the shooting, editing, and publishing all by himself:

The plan must be executable by one person alone; don't assume there's an editor
Give judgments and actions directly; don't write "for reference"
Zero jargon: don't say "A-roll", say "his main talking track"; don't say "ripple", say "the ones after will shift forward with it"

V. 🔴 You must write back after editing (this is the step most easily missed)

After the final cut is published, transcribe the actually broadcast script back and write it into the "Uploaded Version" section of the draft. See the standard at {vault}/00-规则与索引/口播稿输出格式.md §4, "Uploaded Version" below.

Three parts, none may be omitted:

Full text of the uploaded version(code block + line numbers + breath points) — 🔴 only fix proper nouns and typos that ASR misheard,do not change word order, do not smooth out sentences
Difference table(script version vs what he actually said) — use only five types: add / delete / reword / restructure / reorder
Rules distilled from the differences → save to memory

🔴 Why it cannot be skipped: the draft is written, the broadcast version is what he said. Every difference between the two is a vote he cast with his mouth — that is the most honest feedback in this system.

⚠️ Exception: in the difference table, the "reword" columndoes not follow "his is better by default"— speaking is more prone to on-the-spot dulling than writing.Wherever the AI version is sharper, flag it separately and ask.

VI. Absolutely do not

Call the ChatCut tool without first probing the environment (see §0.1)
Do not use apply_script(preview:true) as a read-only preview — it will partially commitwhile reporting an error
Do not reuse old Card ids after changing subtitle font size/box width — pagination will change; ids are content hashes
Do not use tool receipts and metadata as acceptance evidence — see §2b
Do not make up ChatCut features, paths, prices, or version numbers — product facts that change must be checked against the current page of the official Docs
Do not pack multiple checkpoints into a single reply
Do not add background music / B-roll / MG / transitions on your own initiative — if the user did not ask, do not do it
Do not submit video generation without confirmation — it costs money; submit only the items the user checked, one at a time
Do not pick highlight keywords for the user
Do not use local ffmpeg to compress a flattened MP4 as the main deliverable (ffmpeg is only for read-only inspection of source files)
Do not export just because of "a quick cut"

VII. Memory

Self-check before finishing:

Which editing approach did the user reject (record under "correction", e.g. "colloquial noise must not be cleaned")?
Which parameter was empirically verified as effective (record under "effective method", with values)?
Which tool pitfall did you hit(record under "correction",with the verification method: how you found it, how you disproved the wrong guess)?

Write to {vault}/08-技能记忆/aikey-cut/，check for duplicates first。

Do not know the next step → return to /aikey。