Agent skill · Audited from source

App Control

skill · vellum-ai/vellum-assistantMIT licenceAudited 14 Oct 2026 · source 19bc86343b63

★ 1,425pushed 1 day ago
✓SAFEGrade B · 89/100 · 14 Oct 2026 · 19bc863
1,425repo stars · whole vellum-ai/vellum-assistant

Is it safe? Yes — SAFE, grade B (89/100). We read this skill's source at commit 19bc86343b63 and observed filesystem access and no hard-coded secrets. What it does: Drive a specific named macOS app via raw input bypassing the Accessibility tree

Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
Cite

Cite this page

ModelsAtlas. “App Control (vellum-ai/vellum-assistant)”. Retrieved 14 Oct 2026. https://modelsatlas.com/skills/vellum-ai/app-control

Trust audit

Audit 0.4.1 · source read 14 Oct 2026

Trust score89/100✓SAFE B
0 critical0 high0 medium10 low0 info

SAFE means nothing in the source contradicts what the skill says it does. Grade B is shared by 13,985 of 14,816 audited skills.

What the source does

  • FilesystemUndeclared5 observations, not in its own description
  • NetworkNone observedNo outbound calls in the source
  • ShellNone observedNo commands run in the source
  • DependenciesPinnedNothing floats
  • SecretsNone foundNo keys in source

Five audit layers

LayerWhat we checkResultFindings
L0 ProvenanceRepo owner, licence, archive status, activityPASS5
L1 Static sourceSecrets, path traversal, supply chain, hidden codePASS5
L2 InstructionsPrompt injection, authority framing, hidden directivesPASS0
L3 InventoryThe files and declarations match the sourcePASS0
L4 SandboxRun in isolation and watch behaviourSKIPPED why0

Gates applied: no behavioural pass.

Findings (10)

LOWFilesystem / pathfs.traversal · tools/app-control-click.ts:1
F-05 Filesystem / pathCWE-22 · CWE-59Layer L1
import { forwardAppControlProxyTool } from "../../../../tools/app-control/skill-proxy-bridge.js";
Why
No further detail recorded for this rule.
Fix
—
LOWFilesystem / pathfs.traversal · tools/app-control-click.ts:5
F-05 Filesystem / pathCWE-22 · CWE-59Layer L1
} from "../../../../tools/types.js";
Why
No further detail recorded for this rule.
Fix
—
LOWFilesystem / pathfs.traversal · tools/app-control-combo.ts:1
F-05 Filesystem / pathCWE-22 · CWE-59Layer L1
import { forwardAppControlProxyTool } from "../../../../tools/app-control/skill-proxy-bridge.js";
Why
No further detail recorded for this rule.
Fix
—
LOWFilesystem / pathfs.traversal · tools/app-control-combo.ts:5
F-05 Filesystem / pathCWE-22 · CWE-59Layer L1
} from "../../../../tools/types.js";
Why
No further detail recorded for this rule.
Fix
—
LOWFilesystem / pathfs.traversal · tools/app-control-drag.ts:1
F-05 Filesystem / pathCWE-22 · CWE-59Layer L1
import { forwardAppControlProxyTool } from "../../../../tools/app-control/skill-proxy-bridge.js";
Why
No further detail recorded for this rule.
Fix
—

Audit 0.4.1 · source sha 19bc86343b63 · 14 Oct 2026Report an issue or request a re-scan

Provenance

Who published it, and what the verdict is pinned to

  • Publishervellum-ai GitHub organizationOwner of the repository this skill is read from
  • Source foldervellum-ai/vellum-assistant/assistant/src/config/bundled-skills/app-controlThe exact folder we audited
  • LicenceMITAs the repository states it
  • Commit pinned for this verdict19bc86343b63A new audit runs when the source changes
  • Repository statusNot archivedLast push 1 day ago
  • Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

What it tells the agent

Full source on GitHub ↗

Its description: Drive a specific named macOS app via raw input bypassing the Accessibility tree

---
name: app-control
description: Drive a specific named macOS app via raw input bypassing the Accessibility tree
compatibility: "Designed for Vellum personal assistants"
metadata:
  emoji: "🎯"
  vellum:
    display-name: "App Control"
    category: "system"
    feature-flag: "app-control"
    activation-hints:
      - "User explicitly directs the assistant to drive a specific named app via raw input (emulator, game, OpenGL canvas, custom-rendered Electron app)"
      - "User says the macOS Accessibility tree is unhelpful or empty for the target app"
    avoid-when:
      - "Task can be done via the computer-use skill (general macOS UI navigation)"
      - "Task can be done via a CLI / API alternative"
---

This skill exposes the `app_control_*` proxy tools for driving a single
named macOS application via raw input — keyboard, mouse, screenshot — that
bypasses the system Accessibility tree. Use it only when explicitly directed
to a specific app where the AX tree is unhelpful (emulators, games, OpenGL
canvases, custom-rendered Electron apps). For general macOS UI navigation
prefer the `computer-use` skill.

Tools in this skill are proxy tools — execution is forwarded to the connected
macOS client, never handled locally by the assistant.

## Cadence

Take 2-3 actions per turn, then yield with a short narration so the user can
interject. Do not chain long sequences without surfacing what you are doing.

## Always observe before acting

Call `app_control_observe` before your first input action whenever the screen
state matters (e.g. you need to know what is on screen, where a UI element is,
or whether the app is even running). Re-observe after actions that may have
moved the window or changed visibility.

`observe` waits a short settle delay (default ~200ms) before capturing so the
target app and the WindowServer can flush pending input and composite a fresh
frame. If the captured screenshot looks one input behind the latest state
(common with emulators or other slow-feedback apps), pass a larger
`settle_ms`. For static UIs where you just want a quick snapshot, pass
`settle_ms: 0` to skip the wait.

## Input choice

- Prefer `app_control_sequence` over multiple back-to-back `app_control_press`
  calls when sending an ordered batch of presses (e.g. menu navigation,
  repeated movement). Sequence runs in a single round-trip — the target app is
  activated once at the start and the keys are sent serially without any
  window for keyboard focus to drift to another app between presses. Each step
  may carry its own `duration_ms` (hold) and `gap_ms` (pause after).
- Prefer `app_control_combo` over rapid sequential `app_control_press` for
  simultaneous inputs (e.g. cmd+shift+4). `combo` holds every key at once;
  sequential presses interleave key-down and key-up events.
- Use `app_control_type` for literal text into a focused field.

## Coordinate caveat

`app_control_click` and `app_control_drag` use **window-relative** coordinates.
The window may move or resize between observation and click — if you are
uncertain whether the window has shifted, re-observe first.

## App targeting

Use bundle IDs (e.g. `com.example.app`) when possible — they are the most
reliable identifier. Fall back to localized process names if a bundle ID is
unavailable.

## Ending the session

Call `app_control_stop` when you are done. Do **not** auto-quit the controlled
app — `stop` only ends the app-control session, leaving the app running.

Excerpt, first 3,470 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

Install

Pinned to the audited commit 19bc863

Download folder ↗
git clone https://github.com/vellum-ai/vellum-assistant
git -C vellum-assistant checkout 19bc86343b63

The author documents no install command, so this uses the repository directly. Our verdict covers commit 19bc86343b63 only.

Works in

The author has not said which agents this skill supports, and we have not tested it in any.

Audit history

Every audit, every source change

  1. ✓SAFE B 89 / 100

    first audit · source 19bc86343b63

README badge

For maintainers · shows the latest audit

ModelsAtlas: SAFE BMODELSATLASSAFE · B
[![ModelsAtlas: SAFE B](https://modelsatlas.com/skills/vellum-ai/app-control/badge.svg)](https://modelsatlas.com/skills/vellum-ai/app-control/)

The badge shows our verdict and grade for this skill, and links back to this page. It changes when the audit does.

Add this badge to your README

[![ModelsAtlas: SAFE B](https://modelsatlas.com/skills/vellum-ai/app-control/badge.svg)](https://modelsatlas.com/skills/vellum-ai/app-control/)

Questions about App Control

What does the App Control skill do?
An AI Assistant that’s easy to setup, does your work 24/7, knows your preferences and gets better over time.
Is App Control safe to install?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.
What can App Control access on my machine?
The audit observed that it reads or writes files. At least one of those is not mentioned in its own description, which is itself a finding. Secrets in the source: none found.
How current is this page?
The grade is for one exact copy of the source (19bc86343b63), read on 2026-10-14. The repository is watched, and a new audit runs when it changes — this is the first audit.

Sources

Spotted something wrong? Report a correction · Methodology

Keep going in your AI

Opens with this page as the source. We add nothing else to the prompt.