C
cyberspezie/Token-Saver
Claude Code plugin that saves tokens: your Mac or PC does the heavy reading locally (OCR, faces, video scene…
View on GitHub ↗Link checked 2026-11-12 · HTTP 200 · opens the vendor's own site
- Star count
- 0 ★ · cyberspezie/Token-Saver · read from the GitHub search API · (checked 2026-11-13 20:00:50)
- Source
- read directly from its GitHub page · type: open-source project
- Harvest history
- first seen 2026-11-07 · last seen 2026-11-14 · seen 8× · listed by 1 sources
- Score
- score 103 · 1 sources · repo
FAQ (answers come from our records)
Who is it for?
We filed it under these tags: 开源项目. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository
cyberspezie/Token-Saver;first catalogued 2026-11-07;last seen 2026-11-14。There is no official certification or quality rating here — we record sources and dates only.# Token Saver
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
A plugin for Claude Code that saves tokens. Your computer does the heavy reading locally and for free: text recognition, face detection, video scene sheets, file search and file outlines. Claude then reads a one-page summary and at most two images, instead of every screenshot, video frame or long file.
It works silently. There is no status line and no pop-ups: it only changes what Claude reads.
| Without it | With it |
| 6 page screenshots to check a layout: ~6,000 tokens | 1 summary (overflow, broken images, tiny text) + 1 sheet: ~1,700 tokens |
| 12 frames of a 3-minute video: ~15,000 tokens | 1 sheet with one frame per scene + cut list: ~1,700 tokens |
| A 5,500-line HTML file read whole: ~99,000 tokens | A one-page outline: ~1,000 tokens |
| grep hitting a minified CSS line: ~6,000 tokens | The same line clipped: ~70 tokens |
## Install
At the prompt of a Claude Code terminal session:
/plugin install token-saver --marketplace cyberspezie/Token-Saver
Answer y to add the marketplace, then pick the user scope so it works in every project.
From a shell:
claude plugin install token-saver --marketplace cyberspezie/Token-Saver
In the Claude desktop app, where /plugin does not run: click + next to the prompt box, then Plugins, Add marketplace with cyberspezie/Token-Saver, then Add plugin.
## What it does inside Claude Code
Before Claude reads a heavy file, it suggests the local prep step once: an image (a screenshot gets OCR), a video (it gets a scene sheet), or a text file over 60 KB read whole (it gets an outline). If Claude reads the same file again, the read goes through.
Long lines in shell output (minified CSS, JSON) are clipped to 240 characters. Put TOKPREPRAW=1 in a command to get its full output.
A find over the whole home folder is pointed to the system's search index (Spotlight, Windows Search) instead.
Media jobs: when Claude makes an asset for you (a video edit, a cutout, a gif), its temporary files go in one job folder. Once you approve the final version, they are deleted and the conversation is compacted, which frees both disk space and context.
A skill tells Claude which command fits which task.
Code navigation through graphify. The first session after the plugin is installed sets graphify up in the background (uv tool install graphifyy, else pipx, else pip --user) and registers its Claude Code skill. Claude then queries the project's knowledge graph (graphify query, path, explain) before grepping or reading source files, and runs graphify update . after code edits. Build a graph once per project with /graphify .. Once graphify is ready the plugin remembers it and never checks again; a failed install is retried after a day, or at once with tokprep graphify setup --force.
## The tokprep commands
Claude runs these itself; you don't need to.
| Command | What it does locally | What Claude reads |
| tokprep ocr | text recognition | the text |
| tokprep faces | face boxes and capture quality | one line per image + a sheet of face crops |
| tokprep media | size, text, faces, people, scene changes, transcript (macOS 26+) | summary + 1–2 contact sheets |
| tokprep scenes | one frame per scene change (flashes merged), cut list, shot lengths | summary + 1 sheet |
It works silently. There is no status line and no pop-ups: it only changes what Claude reads.
| Without it | With it |
| 6 page screenshots to check a layout: ~6,000 tokens | 1 summary (overflow, broken images, tiny text) + 1 sheet: ~1,700 tokens |
| 12 frames of a 3-minute video: ~15,000 tokens | 1 sheet with one frame per scene + cut list: ~1,700 tokens |
| A 5,500-line HTML file read whole: ~99,000 tokens | A one-page outline: ~1,000 tokens |
| grep hitting a minified CSS line: ~6,000 tokens | The same line clipped: ~70 tokens |
## Install
At the prompt of a Claude Code terminal session:
/plugin install token-saver --marketplace cyberspezie/Token-Saver
Answer y to add the marketplace, then pick the user scope so it works in every project.
From a shell:
claude plugin install token-saver --marketplace cyberspezie/Token-Saver
In the Claude desktop app, where /plugin does not run: click + next to the prompt box, then Plugins, Add marketplace with cyberspezie/Token-Saver, then Add plugin.
## What it does inside Claude Code
Before Claude reads a heavy file, it suggests the local prep step once: an image (a screenshot gets OCR), a video (it gets a scene sheet), or a text file over 60 KB read whole (it gets an outline). If Claude reads the same file again, the read goes through.
Long lines in shell output (minified CSS, JSON) are clipped to 240 characters. Put TOKPREPRAW=1 in a command to get its full output.
A find over the whole home folder is pointed to the system's search index (Spotlight, Windows Search) instead.
Media jobs: when Claude makes an asset for you (a video edit, a cutout, a gif), its temporary files go in one job folder. Once you approve the final version, they are deleted and the conversation is compacted, which frees both disk space and context.
A skill tells Claude which command fits which task.
Code navigation through graphify. The first session after the plugin is installed sets graphify up in the background (uv tool install graphifyy, else pipx, else pip --user) and registers its Claude Code skill. Claude then queries the project's knowledge graph (graphify query, path, explain) before grepping or reading source files, and runs graphify update . after code edits. Build a graph once per project with /graphify .. Once graphify is ready the plugin remembers it and never checks again; a failed install is retried after a day, or at once with tokprep graphify setup --force.
## The tokprep commands
Claude runs these itself; you don't need to.
| Command | What it does locally | What Claude reads |
| tokprep ocr | text recognition | the text |
| tokprep faces | face boxes and capture quality | one line per image + a sheet of face crops |
| tokprep media | size, text, faces, people, scene changes, transcript (macOS 26+) | summary + 1–2 contact sheets |
| tokprep scenes | one frame per scene change (flashes merged), cut list, shot lengths | summary + 1 sheet |
Listed badge (optional)
Copy this line into the README if you want the listing mark. Optional, and it does not affect being listed.
[](https://anclum.com/en/s/cyberspezie-token-saver/)
- Source
- github · https://github.com/cyberspezie/Token-Saver
- Browse by field
- Open source
- added
- today
More like this
Filed under
Every tag opens the full list for that filter. Counts are the live catalogue figures.