菜单
Look at what's actually in an image — describe the scene, subjects, colors, mood, text, composition. Use when the user attaches a photo and asks what's in it, to critique it, or anything that requires seeing the picture content (not just its file facts).
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Build and publish a website. Use whenever the user asks to make, create, build, or publish a website, landing page, dashboard, status page, one-pager, or anything else a human will load in a browser. Covers both the publish recipe (no tunnels, no third-party hosts) and the design discipline that keeps the result from looking like generic AI slop.
Use for Clemson GC curriculum content questions — what a specific course teaches, what skills or projects it includes, what assignments are used, where a topic or learning outcome is covered, and how courses relate to each other pedagogically. This is about course *content*, not degree rules or credit requirements (use gc-advisor for those).
Use for ANY Clemson procurement or purchasing question — what's allowed or required, policy, rules, P-card, requisitions, purchase orders, suppliers, approvals, spending limits, bids/quotes. Use the procurement tools; never answer Clemson procurement from your own knowledge.
Use for Clemson class schedules and instructor scheduling — finding course sections, what an instructor is teaching, section details (time/seats/location), room availability, and academic terms. Use these tools, not your own knowledge.
Use for questions about Clemson Graphic Communications alumni outcomes — where graduates work now, first-job placement, career trajectories, AI exposure of GC occupations, industry/occupation distribution, grad-school rates, and outcome trends by graduation cohort. Data comes from the gc-alumni MCP server (read-only SQLite). For degree rules or course content use gc-advisor or clemson-curriculum instead.
Use for Clemson GC degree rules and academic requirements — degree audit (which requirements has a student met?), credit counts, gen-ed rules, specialty-area rules, prerequisite checking, course planning, and academic regulation lookups. This is about *rules and requirements*, not course content or pedagogy (use clemson-curriculum for what a course teaches, its skills/projects/assignments). Always pulls from the pinned catalog database — never from memory or the live website.
| name | image-vision |
| description | Look at what's actually in an image — describe the scene, subjects, colors, mood, text, composition. Use when the user attaches a photo and asks what's in it, to critique it, or anything that requires seeing the picture content (not just its file facts). |
| allowed-tools | Bash(python3:*) |
When the user attaches an image, the message includes a marker like:
[image: photo.jpg — saved to /workspace/agent/attachments/playground_xxx_0.jpg]
That path is the image file on disk. To actually SEE the picture, run the bundled script with the path and (optionally) a question:
python3 /app/skills/image-vision/scripts/describe.py "/workspace/agent/attachments/playground_xxx_0.jpg" "Is this a strong ad photo? What would you change?"
The script sends the image to a vision model and prints back a description / answer. With no question it just describes the image.
Resolution / cost knob — add --detail:
python3 /app/skills/image-vision/scripts/describe.py "<path>" "Read the text on the sign." --detail high
--detail low — model sees ~512px. Cheapest and fastest; fine for overall
scene, colors, mood.--detail high — image is tiled at full resolution. More tokens (costs more)
but needed for small text, fine detail, or close inspection.--detail auto (default) — the model decides.Use low for quick "what's this?" and high when the user asks about text in
the image or fine details. (--model NAME overrides the vision model.)
Use it for "what's in this photo?", "critique this ad image", "what colors dominate?", "is there text in it?", "does the composition work?" — anything that needs looking at the actual pixels.
For plain file facts (dimensions, aspect ratio, file size) you don't need vision — use the Image info skill, which is faster and needs no network.