Video Chapter Progress Bar

User Flow

First, check if the user has uploaded an SRT file in the current conversation or explicitly provided the path to an SRT file to use. Only subtitles from this source can be read directly. If the current conversation does not contain a user-provided SRT, you must reply: "Please upload the SRT subtitle file for this video first. I'll provide you with a table of chapter suggestions based on the subtitles; after you confirm, I'll create the progress bar." Then wait for the user to provide it. Do not perform file search, subtitle reading, content analysis, or progress bar generation at this point. Do not search the local machine, current workspace, historical projects, download directories, other conversations, or memory for SRT files, nor pick one that seems suitable. Files the user has already provided in the current conversation can be reused without requesting them again. If multiple files are provided without specifying which one to use, ask first. If a specified file is unreadable, ask the user to re-upload or verify the path; do not search for alternative files. Only use built-in sample content when the user explicitly asks to see the style or generate a fixed-duration sample; a simple request like "create a progress bar for me" does not fall under this exception.
Read the entire SRT file, and suggest a number of chapters based on content semantics, topic transitions, video length, and user needs, without setting a fixed upper limit for segments; prioritize the user's specified divisions. Start by stating "I recommend dividing it into X segments," then must list the segmentation suggestions using the table below. Wait for user confirmation before proceeding. When the user requests merging or splitting, update the entire table, then confirm again. Do not replace the table with a plain text list, and do not ask the user to verify timecodes.
After segment confirmation, if the current session has not yet specified the video aspect ratio, you must ask: "What is your video's aspect ratio: 4:3, 16:9, or 3:4? Other ratios can be specified directly." Wait for the answer before proceeding. Do not default to a ratio based on subtitles, old samples, or Playground preview dimensions. If the ratio is already clarified or an original video is available to verify dimensions, use it directly without asking again.
Run scripts/playground_server.mjs of this Skill, open the progressUrl returned by this launch, and display the real dynamic sample. The startup script will automatically reuse the same version of the service, and automatically change the port if it is occupied by other programs or an older version. Do not treat a default port occupation as an installation failure, do not directly reuse an old page from an unverified version, and do not kill unknown processes. Let the user respond with the style name, without requiring them to judge animation implementation or encoding. Do not re-select an already chosen style.
Only process characters when the user selects "walking leading" or "text leading". Read the locally saved personal default character based on references/characters.md. If the user has specified a character for this session, prioritize that selection. Otherwise, if a personal default character exists, use it directly and briefly inform the user, without requesting or generating it again. If no record exists, ask if the user wants to provide a custom image; if not, use a built-in character. A non-reply does not mean "no." After user confirmation of the walking effect for a new custom character, save it to user-data/characters/ within the locally installed Skill, making it the default for next time. If the user explicitly states it is for this session only, do not change the default. Do not ask, read, or save characters when a style without characters is chosen. Personal original images, animation frames, and configurations must not be written to public assets, SKILL.md, Git commits, or GitHub. Preserve local user-data when updating the installation.
Based on the confirmed video aspect ratio, chapters, and style, follow references/rendering.md to save the configuration, dry-run, generate, and check, then deliver a narrow ProRes 4444 MOV. If subtitles are added, removed, or have their speed changed after production, resynchronization is necessary.

Fixed Output Format for Segmentation Suggestions

Segment	Progress Bar Short Title	Content Summary	Starting Dialogue	Ending Dialogue
1	Short chapter name	What this segment is about	Word-for-word quote from SRT of the segment's first line	Word-for-word quote from SRT of the segment's last line

One row per segment; the title is used in the progress bar, the summary helps the user evaluate the segmentation, and the dialogue range helps the user locate it. After the table, ask the user to confirm whether to adopt it, or point out if merging, splitting, or title changes are needed. Original dialogue must not be rewritten as a summary. If too many chapters make the text overflow, explain the actual layout issue and propose shortening titles, adjusting font sizes, or other layout solutions for the user to decide. Do not forcibly compress into 5 segments or arbitrarily merge confirmed chapters to fit a style.

Confirmed information remains valid for the duration of the current session, and is not requested again. Without explicit authorization, do not revise the script, alter subtitles, or edit the original video.

Validated Styles

Soft White Background (soft-white-green): Right-angle horizontal bar, semi-transparent white background, white chapter text and vertical line separators, filling progresses as the segment time advances.
Walking Leading (walking-leading): Uses the same horizontal bar as above, only adds a character walking at the front of the fill. The character is at the same height as the bar, with the body shifted down; at most half a bar height is exposed outside the bar, and the character does not walk on top of the bar.
Dot Line (dot-line): Thin line and dots change color as chapters progress, text is 6 px from the dots.
Short Segments (short-segments): One short line per chapter, filled according to the chapter's time; text is 6 px from the short line.
Text Leading (text-leading): Text color changes as chapters progress, with a character walking at the front; no long bar background color. Small font with no stroke, rendered with 4x precision and single transparency composite.

The RGB fill color of all five styles can be arbitrarily modified. The three new styles support font sizes from 18 to 72 px (based on export pixel count), with defaults of 24, 26, and 25 px respectively; output height automatically adjusts to the text. Green is only the default color; the color selected by the user in the Playground must be written into the final configuration; preview color changes must not result in a green export.

The true source of rules is references/confirmed-rules.md; the registration table is assets/progress/styles.json. Only display and generate validated styles. New styles must be reviewed by the user via actual samples and approved before registration.

Acceptance and Delivery

Scripts check subtitle range, title layout, dimensions, complete decoding, frame count, duration, and transparent pixels. The Agent must also open contact sheets, actual export frames, and dynamic previews to check character loops, cropping, chapter transitions, and colors.
Soft White Background: 1440×64; Walking Leading: 1440×96, where bar height is 64, character max height is 64, and bottom exposure is at most 32. The three new styles have default dimensions of 1440×78, 1440×70, and 1440×96 respectively; for Text Leading, character max height is 64, bottom exposure at most 32. When changing bar height, calculate using the same proportions; do not export the entire canvas by default.
The downloaded file is a narrow ProRes 4444 MOV. Transparent WebM is used for browser preview; reference-qtrle.mov is for internal validation only and is not delivered. Do not replace transparent video with a standard MP4.
Everything except the bar and characters must be transparent. Preview background colors, diagonal lines, or checkerboard patterns must not be baked into the file.
The user has verified that this narrow ProRes 4444 works in CapCut; new files still require checking. Browsers showing normal playback or alpha channel presence does not guarantee support in every CapCut version.
Return the absolute path of the finished file, explaining that it should be placed on the track above the original video, aligned at the start, kept at original speed, stretched to full width, and moved to the top. Clearly mark the sample length; do not present it as a full-length progress bar for the user's video.

For generating a few seconds of independent short animations or chapter transitions, use ll-video-motion; for dubbing, full captions, and continuous material editing, use ll-video-edit.

ll-video-progress