Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add PichurChill/j-can-see --skill j-can-seegit clone --depth 1 https://github.com/PichurChill/j-can-seeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/pichurchill/j-can-see/j-can-see)<a href="https://agentmods.dev/skills/pichurchill/j-can-see/j-can-see"><img src="https://agentmods.dev/badge/skills/pichurchill/j-can-see/j-can-see/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/pichurchill/j-can-see/j-can-see"><img src="https://agentmods.dev/badge/skills/pichurchill/j-can-see/j-can-see.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00140 | $0.04354 |
| Opus 5 | $0.00070 | $0.02177 |
| Sonnet 5 | $0.00028 | $0.00871 |
| Haiku 4.5 | $0.00014 | $0.00435 |
Grade A, and why
j-can-see scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 204 lines — stays where its author put it; the contents beside it link to each section on GitHub.
j-can-see 视觉工具使用指南
本 MCP server 提供 9 个视觉/像素工具,用于主模型无多模态输入能力时的看图任务。 工具会根据任务自动编排,但理解下面的选择逻辑能让你(AI)用得更准。
触发门控:有眼睛就先用眼睛
工具分两类,触发规则不同:
- 视觉识别类(调上游视觉模型):
see_image/locate/inspect/ocr_long—— 你有原生视觉输入(能直接 Read/查看图片)时,默认不要调用这些工具,直接用自己的眼睛。 只有以下情况才用:- 用户明确要求用 j-can-see;
- 原生视觉失效:图片读不了/报错、返回内容降质到不可用、同一问题原生看不出答案;
- 任务超出原生视觉的可靠范围:如两屏以上长图的逐字 OCR(原生读会被降采样丢字)、 需要可靠坐标而目测明显不可信的场景。
- 本地像素类(不调视觉模型):
colors/image_diff/crop/trace/extract_fg—— 不受上述门控限制:它们是取像素级真值的仪器(精确 hex、逐像素差异%),任何模型的 眼睛都给不出这些值,需要时直接用。
不确定自己是否有原生视觉?先试一次原生读图:成功 → 按上面规则走;报错/无图像能力 → 你就是本 server 的目标用户,放心使用视觉识别类工具。
工具选择决策树
按你要回答的问题选工具:
| 问题 | 工具 |
|---|---|
| 这张图是什么 / 说了什么? | see_image |
| 多张图各自是什么?(逐图独立识别) | see_image + each: true |
| 某个东西在哪?(单个目标) | locate |
| 所有同类元素在哪?(每类元素) | inspect |
| 这张超长截图的文字? | ocr_long |
| 把这块裁出来存成文件 | crop |
| 这两张图哪里不同? | image_diff |
| 这里到底是什么颜色? | colors |
| 上下/左右颜色是否一致?渐变哪里断了? | colors(profile / 双 region 对比) |
| 这个图形的矢量形状? | trace |
| 把这个图标抠成透明 PNG | extract_fg |
核心原则:不信任文字描述的像素级事实
视觉模型对颜色、尺寸、坐标的描述是模糊的("浅灰"、"大概在右上")。 凡是要写进代码的像素级事实,用本地工具取真值:
- 颜色 →
colors(返回精确 hex + 候选色匹配) - 差异 →
image_diff(逐像素比较) - 矢量 →
trace(从像素拟合 SVG)
这些工具本身也有边界,用之前先知道:
colors按 5 位量化分桶聚类,适合 UI 纯色;渐变或照片的主色会被打散成 多个小簇,占比数字只能参考。要精确取某处的色,先用region把范围缩到那一块; 比较两处颜色就并行发两次colors(各带一个region)比主色 hex; 要看颜色的空间结构(渐变段 / 接缝 / 断层在哪)用profile: "y" | "x", 它返回均匀段(纯色/渐变)与跳变点的位置和两侧色值。image_diff返回的是 12×12 网格里差异密度最高的格子,不是精确包围盒 —— 一处跨格线的改动会报成好几块。它用来回答"去哪儿看",不是"改动的确切边界"。trace只适合扁平高对比图形;照片和复杂渐变出来的 SVG 没有使用价值。- 放大不增加信息:
crop放大超过原始分辨率的区域时,模型看到的是插值伪影 —— 小尺寸渐变放大后会出现真实的同心圆环带,模型数的"纹理圈数/圆环"可能是在 数伪影,不是原图内容。给模型看的放大 ≤2-3 倍足够;要精确颜色/纹理,用colors+region在原图上取(不经过发给模型的 JPEG 编码,色值不受压缩影响)。
核心工作流:粗到细
多步看图任务遵循「先全局,后局部」:
inspect(扫整体布局,拿到所有元素的坐标)
→ locate(锁定你关注的具体目标,拿精确坐标)
→ see_image(用 region 放大看那个目标的细节)
坐标在工具间是通用的(原图像素):locate/inspect 返回的 x1,y1,x2,y2
直接作为 see_image/crop 的 region 参数传入即可。
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 204 lines · 140 tokens per session scan A d6864b0405a6
j-can-see is a skill published in the GitHub repository PichurChill/j-can-see (0 stars, last pushed 16d ago), licensed MIT. It adds 140 tokens to every session and 4,354 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
html-ppt-zhangzara-retro-zine
A neighborhood zine on the disappearing corner shops — portraits, voices, and what a block loses when they close. Built as a decision-grade story deck for community, local readers.
html-ppt-zhangzara-studio
A photography studio's portfolio-and-rate deck — the signature work, the process, and the packages that win the brief. Built as a decision-grade design craft deck for prospective clients.
motion-frames
A single-frame motion-design composition with looping CSS animations — rotating type ring, animated globe, ticking timer, parallax labels. Renders as a hero video poster you can hand straight to HyperFrames or any keyframe-based exporter. Use when the brief asks for "motion design", "animated hero", "loop", "video…
webgl-halftone-drift
A self-contained WebGL2 hero: a flowing field screened through a rotated halftone dot grid into a duotone print aesthetic; move the cursor to bend the drift.
webgl-holographic-foil
A self-contained WebGL2 hero: thin-film interference over a crushed-foil surface whose palette shifts with the viewing angle; move the cursor to tilt the film.
motion-graphics
A short, design-led motion graphic where motion is the message — kinetic typography, stat count-up, chart/data-viz hit, logo sting / brand lockup, lower-third / callout / social overlay, animated map (highlight regions, connect places, zoom to a location), animated tweet / news-article / headline, webpage / UI…