Vellum Assistant is a personal AI assistant that remembers information about users, learns their preferences, and takes actions across connected apps. It is intended for people who want an assistant that can manage conversations, unfinished work, and proactive notifications over time. The catalogue skills, hooks, instruction, and setting configure or extend how the assistant works.
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 vellum-ai/vellum-assistant --skill screen-annotationgit clone --depth 1 https://github.com/vellum-ai/vellum-assistantWrote 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/vellum-ai/vellum-assistant/screen-annotation)<a href="https://agentmods.dev/skills/vellum-ai/vellum-assistant/screen-annotation"><img src="https://agentmods.dev/badge/skills/vellum-ai/vellum-assistant/screen-annotation/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/vellum-ai/vellum-assistant/screen-annotation"><img src="https://agentmods.dev/badge/skills/vellum-ai/vellum-assistant/screen-annotation.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 | $0.00015 | $0.01895 |
| Opus 5.5 | $0.00006 | $0.00758 |
| Sonnet 5.5 | $0.00003 | $0.00379 |
| Haiku 4.5 | $0.00002 | $0.00189 |
Grade A, and why
screen-annotation 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 1mo 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 — 167 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Drawing on the screen the user is showing you, so they can go and do the thing themselves.
This is the opposite errand from computer use. Nothing here clicks, types or drives anything: the marks are a way of pointing while you talk, for someone who wants to learn where a control is rather than have it operated for them. A mark is drawn clear of what it indicates and never takes the mouse, so what you point at stays visible and clickable the whole time.
Requires a screen share
Marks are drawn on the frame around the surface the user is sharing with the
call. With nothing shared there is nowhere to draw, and screen_point_at
fails saying so. Ask them to share their screen from the call, then point.
Say what to point at
Prefer naming the thing for an arrow.
{"target": "color balance", "caption": "Click this"}.
The name is looked up on the surface itself, which knows where its controls
actually are, and an arrow is drawn at it.
The label, not a description of it. What is matched is the control's own
name. Casing, spacing and punctuation are forgiven, so Color Balance finds
color balance; nothing beyond that is, so "the stabilization button" finds
nothing, because no control is called that. Give the label on its own:
stabilization, Send, Search.
The arrow points at the middle of the control and stops just short, so what you are sending someone to stays visible the whole time.
You are answered with what was drawn and the accessibility name it resolved
to. Some apps expose internal names such as ToolbarExportControl instead of
the visible label. When a returned name identifies the intended control in
the shared image, use that exact name as target; use the visible label in
the caption and speech. Do not invent an internal name or choose an unclear
candidate.
A failed name lookup draws nothing. It means the accessibility name could not be resolved, not that the control is absent from the picture. Read the returned names and retry with an exact name when it identifies the intended control. Do not cycle through invented internal names.
What ships with it
4 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.
- 1mo ago Changed · +9 lines 47f5995f55ce
- 1mo ago Changed · +3 lines 970afca34d2f
- 1mo ago Changed b377560d8bb0
- 1mo ago Changed · +7 lines 77e598ad7c39
- 1mo ago Changed · +67 lines 87594f82c738
- 1mo ago First seen · 81 lines · 15 tokens per session scan A fd5cf44ebdb8
screen-annotation is a skill published in the GitHub repository vellum-ai/vellum-assistant (1,454 stars, last pushed yesterday), licensed MIT. It adds 15 tokens to every session and 1,895 once invoked, about $0.0001 per session on Opus 5.5. 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 2026-09-08.
Other skills, from other repositories
html-artifacts
Author the HTML for a plan artifact, dashboard iframe, or Slack attachment — structure, design plan, available runtime, theming, and craft. Read this before writing HTML for publishartifact, outputiframe, or slackattachhtml.
mermaid-diagram
Use when the user asks for a flowchart, architecture diagram, sequence diagram, ER diagram, state machine, gantt chart, or any visual that fits Mermaid syntax. Hope Agent's chat renders Mermaid blocks natively, so the diagram appears inline.
design-review
Use when a design artifact or revision needs a prepared review packet or an evidence-backed record of a review outcome, requirements, and decisions.
generative-ui
Use when presenting information already in hand as a small, static structured panel in the conversation — status, results, comparisons, or a compact summary — rather than prose, a markdown table, a widget, or an app.
st-widget-builder
Use when the user asks to render, build, visualise, mock up, or show a widget (chart, dashboard, card, table, graph) in the Mewbo Console. Teaches how to delegate the work to the st-widget-builder sub-agent instead of writing HTML or inline code yourself.
frame-composition
Stage the frame on a NodeTool timeline — grids, focal placement, safe areas per aspect ratio, depth layers and parallax, camera moves, and where elements enter and leave. Use when placing titles, lower thirds, shapes or overlays, when a layout must survive 16:9 and 9:16, when adding a push or a parallax move, or when…