zmm-cut

zmm-cut: spoken-video finished-cut editing

It handles "after the shoot". Topic selection, scriptwriting, and review before the shoot belong to /zmm-topic /zmm-script /zmm-review。

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

Here we only work withthe version he actually said. Within that version:

✅ Allowed ❌ Not allowed

Cut (dead shots, repeats, filler words, jargon, cut-off endings) Addany word he didn't say

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

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

⚠️ "This skill doesn't change the script" means no adding words, not that nothing may be touched. Cutting and reordering according to the script rulesis exactly this skill's main battlefield。

✅ Allowed	❌ Not allowed
Cut (dead shots, repeats, filler words, jargon, cut-off endings)	Addany word he didn't say
Reorder(adjust order at the container layer, bring the conclusion to the front)	Smooth sentences, change word order, replace wording
Merge pickups into the original material	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(revised 2026-09-05):
- Host with the ChatCut plugin-version 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, first run a read-only locate (get_active_project) as the server requires to confirm the target project
- None of them / reports signed out → take the references/ChatCut实操.md degraded path in§I: tell them to download + local transcription bypass(all content-layer work requires no ChatCut)
chatcut-talking-head-guide — execution spec for spoken-video editing
Read zmm/references/家族公约.md(if unreadable → state plainly "conventions unreadable, red lines cannot be guaranteed" and stop)
- zmm/references/交互规范.md（🔴 not just read once and done: at wrap-up follow the §IV trio — Recap · Before/After · numbered options for next steps; if information is missing, follow §IV and ask withmultiple-choice questions,one at a time; for inapplicable cases see §V)
Read {vault}/00-规则与索引/口播拍摄基线.md(time/place/attire fixed once and for all)
Read memory {vault}/08-技能记忆/zmm-cut/ + _通用/
🔴 When content is to be changed (not just removing verbal tics), read references/内容层重组.md
🔴 Before the first ChatCut tool call, read references/ChatCut实操.md(three tool pitfalls found in testing)
🔴 After editing, you must return to 口播稿输出格式.md "Upload Version" under §IV — see §V of this document

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

⚠️ Read as needed, don't read it all at once. This skill covers only the "footage shot → final cut" stage; there is no need to read files from the topic-selection layer or the scriptwriting layer.

1. Account default parameters (🔴 already fixed, don't ask the user again)

🟢 Zhan Mingming 2026-08-27 Confirmed.These are default values, not questions to confirm every time. If the user hasn't said to change them, just use them; if the user says otherwise, what the user says this time prevails.

⚠️ The table below shows measured values for "this one account", not general recommendations. If there is config , thenread from config Read; if it can't be read, use the table below andstate that example values are being used. When switching accounts / switching topics, the main workflow still applies,just swap this table。

Item	Default value
Platform	Douyin (vertical 9:16)
Final cut duration	Keep original length — don't cut content just to make it short (the criterion is absolute watch time, not completion rate)
Overall speed	🔴 1.15×(speed up uniformly after the voice editAfterward, not adjusted segment by segment)
Default processing	① Voice edit(remove verbal tics, pauses, repetitions) ② Subtitles
Off by default	B-roll · MG animation · background music — only do them if the user explicitly asks
Pacing style	Tight and punchy

Subtitle specs (🔴 already fixed)

Item	Value
Font	HarmonyOS Sans Bold
Stroke	Light
Shadow	Yes, light
Overall look	Clean and modern
Highlight	★ Yes, color yellow `#FFE20A`
🔴 Who picks the highlighted keywords	The user picks them. Provide candidates when delivering the draft,don't decide for them

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

🔴 talking-head-guide Hard rule:There are dependencies among the multiple processing steps, so they must be finalized in order, and after each major step you must confirm with the user separately; do not bundle multiple checkpoints into one reply. If something upstream changes, everything downstream must be redone (writing subtitles against the pre-speed-up timeline = all 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
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 Material inventory and merging (added 2026-09-05)

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

Firsttranscribe each one, read them separately,don't splice them together right away
State clearlywhich one is the main material and which is the pickup, and which segment the pickup shouldcover
🔴 Once a pickup is added, you must do a "verbatim repetition scan" — see references/内容层重组.md §2.2
Mergein the new order, not by appending the pickup to the end

① Ingest material into the project

First list_projects / create_project / target_project set up the project,then do the rest
Once the project is createdimmediatelysurface the editor (preview_start), so the user can watch the progress
Material can come in three ways: the user uploads directly in the editor · a local path is given in the session (via asset-import) · the user has already uploaded it (first browse_assets look for it,don't just say it can't be found）
You can start working as soon as the transcript is out, without waiting for the raw bytes to finish uploading (A-roll only needs the transcript)

② Voice editing (the main battleground of this skill)

🔴 This step has two layers; don't mix them: Mechanical layer(fixed verbal tics, batch pauses) doesn't require understanding meaning; tools can do it; Content layer(what to keep, what to delete, what to reorder)requires understanding meaning, the criteria are in references/内容层重组.md。

⚠️ Doing only the mechanical layer = just wiping the footage clean, not improving it. When the user says "cut it a bit", they usually mean both layers,but every deletion or change in the content layer must have a clear reason stated。

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

read_script → read through timeline.mdto see the structure
First run mechanical cleanup（clean_script): handle only fixed verbal tics (uh/um/mmm) and long pauses
🔴 After running, you must re-read the refreshed timeline.md — what you read before is now stale
Then docontent layerjudgment — go by references/内容层重组.md 's five criteria: identify dead shots · verbatim repetition scan(mandatory for pickups) · reasoning chains must not be reordered for suspense · the opening must have both a hook and resonance · before deleting, ask "will it still be clear after deleting this?"
apply_script lands on the timeline
Read it back and check:seam logic(is a bridge missing) · consistency before and after · whether relative time still holds · whether the ending is complete · whether the pauses are too tight

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

Keep	Why
Colloquial noise(ne / a / ha / reduplication / slip-of-the-tongue repetition)	Rule Every sentence being neat = the audience instantly hears it's being read from a script. ⚠️ Generic "remove verbal tics" will clear these out too — keeping one every 5–8 lines is by design, not a mistake
Inserted lines inviting self-reflection("think about it carefully", "I don't know if you've ever had moments like this")	Rule It carries an action instruction, not a filler word
Qualifiers(mostly / a lot of the time / maybe that's what)	Rule: must not be changed into an absolute statement

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

Pause handling

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

③ Overall speed-up 1.15×

🔴 Must be done after the audio edit is finalized, if the order is reversed all subtitles must be redone. Uniform speed-up,not per-segment speed adjustment — per-segment adjustment makes the intonation speed up and slow down unpredictably.

④ Subtitles + highlights

According to talking-head-guide 's captions specification, apply the fixed spec in §1.

🔴 I must not decide the highlight 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 closing sentences)
List them for the user to choose, clearly indicating the second and the original sentence
Apply highlights only after the user has checked them

⚠️ 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.

⑤ Delivery

🔴 The default delivery 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", "make a version" do not count as export intent.

2b. 🔴 Acceptance discipline: metadata lies (added 2026-09-05, tested three times)

Line count, text, and image can only be judged by full-size rendered frames.

fooled	actual
Subtitle interface reported "one line, no overflow"	rendered as two lines
In the multi-grid thumbnail the text looked wrong	a full-size single frame proved it was a misreading
Tool receipt reported "cleaned up 1 word"	the timeline was actually 4.9 seconds shorter

After each major step, always verify two things：

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

⚠️ Thumbnails are only used for the rough 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 the time of shooting. Editing only deletes, never adds,don't add words for him, smooth out sentences, or change word order
Visual information points	Outfits / desk setup / background / wall decorations are all information points.When editing, don't cut out shots that carry information
Don't include AI screenshots	the visuals of the segment talking about AIDon't include AI interface screenshots — it will narrow the audience (`不要变成工具号`）
Where to put big text	In the script's "points that need handling", lines marked ⏸ are the big-text lines,they must line up during editing
Blur amounts	Any frame showing backend data / payments / customer info must be blurred

Four, Who it's for

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

The plan must be executable by one person; 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 "the main track where he's talking"; don't say "ripple", say "the stuff after it moves forward with it"

Five, 🔴 Must backfill after editing (this step is the easiest to miss)

After the final cut is published, transcribe the text actually broadcast back and write it into the script's "upload version" section. For the spec, see {vault}/00-规则与索引/口播稿输出格式.md "upload version" under §Four.

Three parts, none can be missing:

Full text of the upload version(code block + line numbers + breath points) — 🔴 only fix proper nouns misheard by ASR and typos,don't change word order, don't smooth out sentences
Diff table(script version vs what he actually said) — use only five types: add / delete / reword / restructure / reorder
Rules distilled from the diffs → add to the library

🔴 Why it can't be skipped: the script 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 diff table, the "reword" columndoes not follow "his version 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.

Six, Never do

Don't call ChatCut tools before probing the environment (see §0.1)
Don't use apply_script(preview:true) as a read-only preview — it errors while alsopartially committing
Don't reuse old Card ids after changing subtitle font size/box width — it will repaginate; ids are content hashes
Don't use tool receipts and metadata as the basis for acceptance — see §2b
Don't make up ChatCut features, paths, prices, or version numbers — product facts that change must be checked on the current page of the official Docs
Don't pack multiple checkpoints into one reply
Don't add background music / B-roll / MG / transitions on your own initiative — if the user didn't ask, don't do it
Don't pick highlight keywords for the user
Don't use local ffmpeg to compress a flattened MP4 as the main deliverable (ffmpeg is only for read-only inspection of source files)
Don't export just because of "a quick edit"

Seven, Memory

Self-check before finishing:

Which editing approach did the user reject (record as "correction", e.g. "colloquial noise must not be cleaned")?
Which parameter has been empirically verified as effective (record as "effective method", with numerical values)?
Which pitfall of the tool was encountered(record as "correction",with evidence-gathering method: how it was discovered, how the wrong guess was disproven)?

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

Don't know the next step → go back to /zmm。