Video Report Agent skill
The video-report skill turns a video transcript into a self-contained HTML report with chapters and diagrams tied to the source.
- 315 ★ on the repository
- Last commit
Extended access
Why this level
- The full CLI runs local Pi commands with user rights and downloads media through yt-dlp.
- Cloud recognition uploads audio, and the report model receives the transcript.
- Model keys live in `.env`.
Install
- Works in
Manual install
git clone https://github.com/imexlovery/video-report-agent.gitThe first README option with a command. The skill sits in `src/video_report_agent/skills/video-report/`; a ready Skill ZIP is in the repository releases.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The full CLI runs local Pi commands with user rights and downloads media through yt-dlp.
- Cloud recognition uploads audio, and the report model receives the transcript.
- Model keys live in `.env`.
This is third-party code. Review the repository files before installing.
What it does
The main repository artifact, the `video-report` skill, leads an agent from a full transcript to a readable report. The agent reads the whole source, splits it into chapters by meaning, draws diagrams for mechanisms and steps, drops promotion and keeps numbers, conditions and caveats from the author. The output is a single `report.html` with inline styles. There are two modes: `standard` for full reading and `brief` for a quick digest. Beyond the skill the repository has a Python `video-report` CLI that fetches subtitles or recognizes speech, runs the agent and exports a long PNG.
Who it is for. People who work through long videos and lectures: researchers, editors and students who need a skimmable report checkable against the source.
Good fit when
- A two-hour talk has to become a report with chapters, figures and the author conditions.
- A transcript already exists as Markdown, TXT or SRT and a reading page remains to be built.
- To grasp a video quickly, a brief version of 1600–2200 characters is needed.
- Bilibili videos become reports through the CLI with automatic subtitle retrieval.
Not a fit when
- Speech recognition inside the skill is needed: it contains no ASR and does not configure a model for you.
- The source is a closed or private video: the built-in entry covers public Bilibili.
- A retelling without checking is wanted: the report is only as good as the transcript.
Example request
Read `video-report/SKILL.md`, `standard` mode, and build a self-contained `output/report.html` from `transcript.md`.Limitations
The skill has no ASR, so a ready transcript is needed; report quality depends on the transcript and the model. Documentation is mostly Chinese, with `README.en.md` available. The full CLI targets public Bilibili videos up to 5 hours, and native Windows is unverified. Cloud recognition uploads audio to a third-party service, and `PI_TASK_ISOLATION=local` gives Pi tools user-level rights. By default the analysis uses text only: information from the picture may be lost.
How to disable. Delete the `video-report/` folder from the agent skills directory. If you installed the CLI, delete the cloned repository and the `.env` file with keys.
FAQ
What is the output?
A single `report.html` with inline styles. Open it in a browser; the CLI makes a long PNG if installed.
Which transcript formats fit?
Markdown, TXT and SRT, and any long text works. Title, author, links and timestamps are optional.
How does `standard` differ from `brief`?
`standard` keeps explanations, grounds and examples. `brief` gives a quick digest: a target of 1600–2200 characters with a cap of 2400.
Do I need the full Python pipeline?
No. The skill needs only a coding agent and the `video-report/` folder; the pipeline is for running from a Bilibili link.
Where does the CLI get the transcript?
It tries author subtitles, Bilibili automatic `ai-zh` subtitles and speech recognition in turn. The `--sources` flag picks the source.
Discussion
No questions or reviews yet.
Related
SkillsSkill that turns a video into tall subtitle collages: crops burned-in subtitles or draws reviewed lines onto real frames
A skill that makes video in pure code: narrated explainers and website promos in 35 visual styles, in 16:9 and 9:16.
Twelve skills teach an agent to read YouTube through the TranscriptOut service: transcripts, search and channels in one HTTPS call
watch-cli is a skill and CLI that hands an agent the frames and transcript of any social video so it can read the clip itself.
A skill that turns a Bilibili, Douyin or YouTube video link into timestamped notes through local transcription
A Codex skill that breaks a reference video into timed shots and writes a prompt reusing its story mechanics for your own product.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Video Report Agent (this card) | no | 315 | |
| YouTube Skills | no | 275 | skills around the YouTube transcript service; here the input is a ready transcript and the result is an HTML report with chapters and diagrams |
| bilibili-cli | no | 1,152 | a CLI for reading Bilibili videos, subtitles and feeds; Video Report Agent turns the retrieved text into a report |
README in short
The README gives two paths: the skill for your own agent and the full `video-report generate` CLI for Bilibili videos. The skill can be downloaded as a `Skill ZIP v0.5.1` archive.