alibabacloud-adb-highlight-editing

name	alibabacloud-adb-highlight-editing
description	Use when a user provides Alibaba OSS pre-signed media URLs and asks for public highlight extraction, continuous editing (顺剪), or story-cut editing (混剪) with local downloads.

ADB Highlight Editing

Workflow

1. Identify the operation.

   Intent	Command	API mode
   高光提取、提取片段	extract	extraction
   顺剪、按原剧情连续剪	continuous	CONTINUOUS
   混剪、组合高光剧情	story-cut	STORY_CUT

   Ask only when the mode is ambiguous.

2. Require the user to provide:

   • the public service origin (scheme, host, and optional port); the skill appends the fixed /api/v1/operators path itself;
   • one complete public Alibaba OSS pre-signed HTTP(S) file URL per input.

   Do not use a default API address. Do not accept a bucket, prefix, unsigned URL, arbitrary HTTP URL, or an OSS internal URL. Ask for complete URLs when any input is missing or invalid.

3. Convert each ordered media URL into one --media argument. Maximum: 20. The CLI validates a regional OSS host, a file path, and OSS V1 or V4 pre-signature parameters. It then adds -internal to the OSS hostname while preserving the path and complete query string.

   Preserve story order. Treat the input media list as a candidate pool; an episode range does not require every input to appear in the output. Only pass --require-all-media when the user explicitly requires every input episode to appear in each output. "Use/process episodes 01–05" does not enable it.

4. Run scripts/highlight_service.py relative to this file. Pass the user's public service origin explicitly; the CLI appends /api/v1/operators.

   python "<skill-directory>/scripts/highlight_service.py" story-cut \
     --api-base-url 'http://service.example:8000' \
     --media '<complete-public-oss-pre-signed-url-for-episode-01>' \
     --media '<complete-public-oss-pre-signed-url-for-episode-02>' \
     --highlight-description '围绕核心冲突，保留铺垫、升级、转折和结果' \
     --target-duration 90 --output-count 1 \
     --output-dir "$PWD/highlight-results/story-cut"
   

   Use the command's --help. Read references/api.md for duration, hook, output-count, all-media, and error details.

5. Let the command poll to terminal status. Keep its session alive. Never resubmit because it is slow. GET retries are automatic; POST is not.

6. Report task ID, elapsed time, and downloaded paths from summary.json. Also retain redacted result.json. Explicitly report an empty result.

Quick Reference

Need	Arguments
10–15s highlights	extract --min-duration 10 --max-duration 15 --speed-min 0.8 --speed-max 1.2
Continuous edit	continuous --target-duration 120 --opening-duration 10
Story cut (candidate pool by default)	story-cut --target-duration 120 [--require-all-media only if explicit]
Submit only	--submit-only
No rendered files	--no-export
Check task	status --kind extraction|editing --task-id TASK_ID --api-base-url URL
Retry downloads	download --operation extract|continuous|story-cut --task-id TASK_ID --api-base-url URL --output-dir PATH

Failure Rules

• Invalid media URL: ask for a complete, unexpired public OSS pre-signed file URL. The CLI validates its structure; OSS validates the signature itself.
• FAILED: report error and task ID; do not resubmit without user direction.
• Timeout: preserve task ID and use status, not POST.
• Download failure: use download on the same successful task.
• Never expose OSS URL query parameters in chat or retained result files.