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 watch-togethergit 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/watch-together)<a href="https://agentmods.dev/skills/vellum-ai/vellum-assistant/watch-together"><img src="https://agentmods.dev/badge/skills/vellum-ai/vellum-assistant/watch-together/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/watch-together"><img src="https://agentmods.dev/badge/skills/vellum-ai/vellum-assistant/watch-together.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Privilege Escalation · line 237 Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.Fix: Avoid sudo/root unless strictly required. Prefer least-privilege patterns. If elevation is needed, document the justification and scope.
- medium Privilege Escalation · line 243 Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.Fix: Avoid sudo/root unless strictly required. Prefer least-privilege patterns. If elevation is needed, document the justification and scope.
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 | $0.00053 | $0.02043 |
| Opus 5 | $0.00026 | $0.01022 |
| Sonnet 5 | $0.00011 | $0.00409 |
| Haiku 4.5 | $0.00005 | $0.00204 |
Grade B, and why
watch-together scanned grade B with 1 finding 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 9d 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.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
sudo apt install python3 ffmpeg mpv How it starts
The opening of the file, as written. The whole thing — 259 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Watch Together
Real-time co-watching. An editor model watches playback continuously and wakes the assistant only at moments worth reacting to. Each wake includes dialogue and story-critical frames. Wakes arrive through signal files, so do not poll.
Choose a mode
Source mode (preferred)
watch-file.py plays a local media file in mpv and reads that file directly.
It does not capture the screen or require a loopback audio driver. Frames come
from the source at full quality. Sidecar .srt subtitles or the first embedded
text subtitle track provide exact dialogue. Without subtitles, the editor
transcribes audio. Pause stops the flow and seeks resynchronize it.
Use source mode whenever the user has a local media file. The bundled script does not accept stream URLs.
Screen mode (fallback)
capture_live.py records supported desktops through ffmpeg and feeds completed
segments to the same editor pipeline.
- macOS: Uses AVFoundation. It auto-detects BlackHole for audio and records video only when BlackHole is unavailable.
- Windows: Uses ffmpeg
gdigrab. It records video only by default. System audio requires an existing DirectShow playback source such as Stereo Mix or a configured virtual audio cable, passed explicitly as the audio device. - Linux: Uses X11 capture and
$DISPLAY. It records video only by default. System audio requires an existing PulseAudio or PipeWire-Pulse monitor source passed explicitly. Native Wayland screen capture is not supported by this script.
Screen capture cannot bypass DRM or protected-video blanking. If browser video is black in the recording, explain that screen mode is unsupported for that content. Do not claim the fallback works around content protection.
How it works
- Playback is divided into 60-second windows.
editor.pyreviews each window, selects story-critical frames, collects dialogue, and decides whether to wake now or hold a developing moment.- A wake extracts the selected frames at 720p and sends a
[WATCH]message to the active conversation. - The assistant looks at the frames and dialogue and reacts only when useful.
What ships with it
6 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.
- 9d ago First seen · 259 lines · 53 tokens per session scan B ce54f1bd1ab7
watch-together is a skill published in the GitHub repository vellum-ai/vellum-assistant (1,234 stars, last pushed today), licensed MIT. It adds 53 tokens to every session and 2,043 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 1 finding (asks for root). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
lictor-security-check
Pre-release security audit for ANY project — AI-built or hand-written, web or mobile. Scans the codebase for the full range of real-world risks that get apps breached: leaked API & AI-provider keys, exposed configs/secrets, broken auth & access control (IDOR), injection (SQL/XSS/command), SSRF, open databases & cloud…
lictor-rotate
Walks the user through rotating a leaked API key — step by step, provider-specific. Knows the exact URL to visit, the exact button to click, and how to verify the rotation worked. Supports Stripe, OpenAI, Anthropic, Google Cloud / AI Studio, GitHub, AWS, Slack, Supabase, Firebase, Postmark, and generic OAuth providers.
lictor-explain
Takes any security finding, error message, or jargon-heavy security advice and explains it in plain English. Use this when someone is confused by what /lictor-security-check found, or when they got a security warning from another tool and don't understand it.
lictor-fix-it
Applies the fixes recommended by /lictor-security-check, with the user's explicit permission for each change. Walks through findings one at a time, shows the proposed change, gets approval, applies, runs tests if available, and moves on. Some fixes (rotating leaked credentials) require the user to act outside Claude …
continual-learning
Nightly refinement of an existing per-repo review-style prompt using this reviewer's own finding outcomes. Read confirmed (resolved-by-commit / thumbs-up) and dismissed (thumbs-down) findings, promote the bug patterns the team actually fixes, demote the false-positive patterns, reconcile against the current prompt…
ha-logs
A read-only troubleshooting skill for querying Hope Agent’s local SQLite databases, which store logs, conversations, and background-job status.