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 TIanle-art/agent-vision-skill --skill visiongit clone --depth 1 https://github.com/TIanle-art/agent-vision-skillWrote 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/tianle-art/agent-vision-skill/vision)<a href="https://agentmods.dev/skills/tianle-art/agent-vision-skill/vision"><img src="https://agentmods.dev/badge/skills/tianle-art/agent-vision-skill/vision/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/tianle-art/agent-vision-skill/vision"><img src="https://agentmods.dev/badge/skills/tianle-art/agent-vision-skill/vision.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 | $0.00121 | $0.01272 |
| Opus 5 | $0.00060 | $0.00636 |
| Sonnet 5 | $0.00024 | $0.00254 |
| Haiku 4.5 | $0.00012 | $0.00127 |
Grade A, and why
vision 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 8d 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 — 91 lines — stays where its author put it; the contents beside it link to each section on GitHub.
识图能力
底层模型不具备原生识图能力。遇到图片时，不要用 Read 工具，改用本项目脚本 vision.js：
node "scripts/vision.js" "<图片路径>" "用中文描述这张图片"
触发场景
- 用户分享图片路径（本地或网络 URL）
- 消息中出现 "Saved attachments:" 并列出图片
- 用户要求分析、描述、识别图片内容
多图
用户一次发多张图片时，必须逐张处理全部图片，拿到所有描述后再回复：
node "scripts/vision.js" "图片1路径" "图片2路径" "用中文描述每张图片"
拿到全部描述后再组织回复，不得只处理第一张。
网络图片
用户发来的是图片 URL 时，直接传即可（自动识别，也可用 --url 显式指定）：
node "scripts/vision.js" "https://example.com/a.png" "用中文描述这张图片"
切换模型
用户说"换增强视觉模型"或"换 plus"时：把视觉模型改为 qwen3-vl-plus。
用户说"换回 flash"时：改回 qwen3-vl-flash。
用户指定其他模型名时，按其要求填写。
改哪里：先看 vision.js 同目录有没有 .env 文件——有就改 .env 里的 VISION_MODEL（环境变量优先级高于代码），没有 .env 才改 vision.js 顶部模型配置区的 MODEL。（skill 安装形态下 .env 放在 skill 根目录，同样优先改它）
配置
第一次使用前，把 .env.example 复制为 .env（放在 vision.js 同目录；skill 安装形态放 skill 根目录）并填入 API Key：
DASHSCOPE_API_KEY=sk-xxx
没有 Key 时：提示用户去 https://bailian.console.aliyun.com/ 注册申请（新用户有免费额度），拿到 Key 再继续。
常见错误
文件不存在: <路径>：图片路径写错，先确认文件真实存在API 401/403：API Key 错误或失效API 404：模型名错误或未开通，或 BASE_URL 服务地址错误图片过大：超过 7MB（macOS 可用sips -Z 2000 <图片> --out <新文件>压缩，Windows 用画图/图片工具缩放；或改.env的VISION_MAX_IMAGE_MB）图片分辨率过大：超过 8K（macOS 可用sips -Z 7680 <图片> --out <新文件>压缩，Windows 用画图/图片工具缩放）不支持 GIF 动图：gif 需先转 jpg/png（macOS 可用sips -s format jpeg <图片> --out <新文件>.jpg，Windows 用图片工具转存）- 失败会自动重试（429/5xx/超时，最多 2 次）
粘贴图路径未知时（消息里只有 [Image N]，无文件路径）
部分 agent（如 opencode）粘贴图片时不写临时文件，而是把 base64 存进本地 SQLite 数据库（opencode: ~/.local/share/opencode/opencode.db 的 part 表），消息里只有 [Image N] 占位。不要按临时目录"最新文件"猜路径（clipboard_image.png 之类可能是旧会话残留，会读错图）。直接让脚本自动定位：
node "scripts/vision.js" --locate "用中文描述这张图片"
--locate 会从已知存储位置（当前支持 opencode 数据库）恢复最新图片附件、解码并直接识别；找不到时报错提示手动传路径。
多会话并发怕取错图时，可用环境变量限定会话（当前会话 ID 可通过查询自己的工具调用记录得到：SELECT DISTINCT session_id FROM part WHERE data LIKE '%<自己的工具名>%' ORDER BY time_created DESC LIMIT 1;）：
VISION_OPENCODE_SESSION=<会话ID> node "scripts/vision.js" --locate "用中文描述这张图片"
识别结果明显与用户描述的图不符时，先查数据库而不是再猜路径。
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 8d ago First seen · 91 lines · 0 tokens per session scan A c601349e189e
vision is a skill published in the GitHub repository TIanle-art/agent-vision-skill (2 stars, last pushed 1mo ago), licensed MIT. It adds 121 tokens to every session and 1,272 once invoked, about $0.0006 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
fixing-prompt
Prompt: Prompt Refinement and Optimization.
llm-cost-advisor
WHAT — Recommend the most cost-effective LLM provider for a given task type. Shows estimated cost per run across available providers and integrates with devcompanion llm-status to show what is actually available.
dbt-validation
HOW — Run dbt checks as documented in the target repo (parse, compile, test, selective run). Does not configure Snowflake accounts or change cloud security.
supamem
Project-agnostic dual-memory tooling — semantic + structural memory for AI coding agents.
boost-prompt
Interactive prompt refinement workflow: interrogates scope, deliverables, constraints; copies final markdown to clipboard; never writes code. Requires the Joyride extension.
deapi
AI media generation via deAPI. Transcribe YouTube/audio/video, generate images from text, text-to-speech, OCR, remove backgrounds, upscale images, create videos, generate embeddings. 10-20x cheaper than OpenAI/Replicate.