Live survey304 models · 14 providers · 23,592 auditsAudited implement · SAFE →What changedHow we verifyllms.txt

Agent skill · Audited from source

agent-plugin

skill · ChatCut-Inc/agent-pluginApache-2.0 licenceAudited 12 Nov 2026 · source c8e8b0479de1

★ 1,030pushed todayversion 2.0.4
✕BLOCKGrade D · 66/100 · 25 findings · c8e8b04
1,030repo stars · whole ChatCut-Inc/agent-plugin

Is it safe? No — BLOCK, grade D (66/100). We read this skill's source at commit c8e8b0479de1 and observed network and shell access and no hard-coded secrets. The audit raised 25 findings, 10 critical or high. What it does: Install, open or connect ChatCut Desktop in Antigravity (安装 ChatCut), or start editing or recreating a video with it: edit or cut a video, add captions or subtitles, clean up a talking-head recording, add B-roll, music…

Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
Cite

Cite this page

ModelsAtlas. “agent-plugin (ChatCut-Inc/agent-plugin)”. Retrieved 12 Nov 2026. https://modelsatlas.com/skills/chatcut-inc/agent-plugin

Blocked. The page stays public so you can check this skill if someone recommended it to you. The audit found 10 critical or high findings, starting with “Prompt injection” in antigravity/skills/asset-import/SKILL.md:49. Do not install it without reading them.

Trust audit

Audit 0.4.1 · source read 12 Nov 2026

Trust score66/100✕BLOCK D
4 critical6 high9 medium7 low1 info

BLOCK means the audit found something that could harm you or your machine. Grade D is shared by 749 of 18,389 audited skills.

What the source does

  • FilesystemNone observedNo reads or writes in the source
  • NetworkDeclared6 observations
  • ShellDeclared5 observations
  • DependenciesPinnedNothing floats
  • SecretsNone foundNo keys in source

Five audit layers

LayerWhat we checkResultFindings
L0 ProvenanceRepo owner, licence, archive status, activityWARN6
L1 Static sourceSecrets, path traversal, supply chain, hidden codeWARN6
L2 InstructionsPrompt injection, authority framing, hidden directivesFAIL13
L3 InventoryThe files and declarations match the sourcePASS0
L4 SandboxRun in isolation and watch behaviourSKIPPED why0

Gates applied: critical finding, instruction override, no behavioural pass, undeclared transfer.

Findings (25)

CRITICALPrompt injectionprompt.transfer_instruction · antigravity/skills/asset-import/SKILL.md:49
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
"<bundled-or-global-node>" <this-skill-dir>/scripts/upload-media.mjs --token <token> --endpoint <endpoint> /path/to/source.mp4
Why
an instruction to move sensitive data to an outside destination
Fix
remove; a skill never needs the user's secrets off the machine
CRITICALPrompt injectionprompt.transfer_instruction · antigravity/skills/chatcut-plugin-basics/SKILL.md:60
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
2. **Authentication required:** Open Agent settings (**Cmd+,** / **Ctrl+,**) → Customizations → **Authenticate** next to the ChatCut server → finish ChatCut sign-in in the browser → paste the code bac
Why
an instruction to move sensitive data to an outside destination
Fix
remove; a skill never needs the user's secrets off the machine
CRITICALPrompt injectionprompt.transfer_instruction · claude-directory/skills/chatcut-editing-basics/SKILL.md:3
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
description: Start here for any ChatCut video editing task with the ChatCut connector (the `chatcut` MCP server). Covers signing in, finding or creating a project, getting the user's media in through
Why
an instruction to move sensitive data to an outside destination
Fix
remove; a skill never needs the user's secrets off the machine
CRITICALPrompt injectionprompt.transfer_instruction · claude/skills/asset-import/SKILL.md:49
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
node <this-skill-dir>/scripts/upload-media.mjs --token <token> --endpoint <endpoint> /path/to/source.mp4
Why
an instruction to move sensitive data to an outside destination
Fix
remove; a skill never needs the user's secrets off the machine
HIGHPrompt injectionprompt.credential_read · claude/skills/video-translation/SKILL.md:23
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
authentication. Do not read passwords, API keys, or OAuth credentials from the
Why
asks the agent to read credentials
Fix
—
HIGHPrompt injectionprompt.hide_from_user · antigravity/skills/product-help/SKILL.md:85
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
the Web release is live, do not tell the user to install an update.
Why
asks the agent to act without the user's knowledge
Fix
—
HIGHPrompt injectionprompt.hide_from_user · codex/skills/product-help/SKILL.md:85
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
the Web release is live, do not tell the user to install an update.
Why
asks the agent to act without the user's knowledge
Fix
—
HIGHPrompt injectionprompt.hide_from_user · docs/claude-code-web-install.md:200
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
Completion gate: do not tell the user ChatCut is installed or setup is
Why
asks the agent to act without the user's knowledge
Fix
—
HIGHPrompt injectionprompt.hide_from_user · kimi/skills/product-help/SKILL.md:79
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
the Web release is live, do not tell the user to install an update.
Why
asks the agent to act without the user's knowledge
Fix
—
HIGHPrompt injectionprompt.hide_from_user · qwen/skills/app-promo-guide/SKILL.md:90
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
Preset reference assets inside `<reference_assets>` **must not be registered in the asset library**. They are internal production references. Pass them directly by URL during generation, through the r
Why
asks the agent to act without the user's knowledge
Fix
—
MEDIUMNetwork egressnet.raw_ip · codex/skills/asset-import/scripts/serve-local-media.mjs:181
F-04 Network egressCWE-200 · CWE-319Layer L1
const base = `http://127.0.0.1:${addr.port}/${routePrefix}`;
Why
No further detail recorded for this rule.
Fix
—
MEDIUMNetwork egressnet.raw_ip · grok/skills/asset-import/scripts/serve-local-media.mjs:165
F-04 Network egressCWE-200 · CWE-319Layer L1
const base = `http://127.0.0.1:${address.port}/${routePrefix}`;
Why
No further detail recorded for this rule.
Fix
—
MEDIUMObfuscation / stealthobf.base64_blob · kimi/kimi.plugin.json:35
F-07 Obfuscation / stealthCWE-506 · CWE-94Layer L1
"iconUrl": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAEAAAABACAYAAACqaXHeAAAABGdBTUEAALGPC/xhBQAAACBjSFJNAAB6JgAAgIQAAPoAAACA6AAAdTAAAOpgAAA6mAAAF3CculE8AAABAmVYSWZNTQAqAAAACAAHARIAAwAAAAEAAQAAAR
Why
No further detail recorded for this rule.
Fix
—
MEDIUMPrompt injectionprompt.conditional_escalation · claude-directory/skills/chatcut-editing-basics/SKILL.md:42
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
3. When the user says the upload finished, find the new assets with
Why
No further detail recorded for this rule.
Fix
—
LOWInventory / provenanceinv.hidden_file · kimi/.app.json
F-13 Inventory / provenanceCWE-1104Layer L0
.app.json
Why
hidden member outside the usual dotfiles
Fix
review its purpose
LOWFilesystem / pathfs.traversal · tests/cursor-marketplace-split.test.mjs:90
F-05 Filesystem / pathCWE-22 · CWE-59Layer L1
assert.equal(await readlink(cursorSkill), `../../grok/skills/${skill}`);
Why
No further detail recorded for this rule.
Fix
—
LOWNetwork egressnet.raw_ip · docs/claude-code-web-install.md:70
F-04 Network egressCWE-200 · CWE-319Layer L1
`HTTPS_PROXY` and `HTTP_PROXY` to `http://127.0.0.1:7897`, rerun the failed
Why
No further detail recorded for this rule.
Fix
—
LOWNetwork egressnet.raw_ip · tests/serve-local-media.test.mjs:78
F-04 Network egressCWE-200 · CWE-319Layer L1
assert.equal(output.files[filename], `http://127.0.0.1:${output.port}/${output.token}/${encoded}`);
Why
No further detail recorded for this rule.
Fix
—
LOWPrompt injectionprompt.credential_read · claude-directory/README.md:17
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
never read credentials from local files or environment variables. See the
Why
asks the agent to read credentials
Fix
—
LOWSupply chainprompt.pipe_to_shell · README.md:74
F-08 Supply chainCWE-829 · CWE-1357Layer L2
curl -fsSL https://chatcut.io/install.sh | sh
Why
No further detail recorded for this rule.
Fix
—

Audit 0.4.1 · source sha c8e8b0479de1 · 12 Nov 2026Report an issue or request a re-scan

Provenance

Who published it, and what the verdict is pinned to

  • PublisherChatCut-Inc GitHub organizationOwner of the repository this skill is read from
  • Source folderChatCut-Inc/agent-pluginThe exact folder we audited
  • LicenceApache-2.0As the repository states it
  • Commit pinned for this verdictc8e8b0479de1A new audit runs when the source changes
  • Repository statusNot archivedPushed today
  • Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

What it tells the agent

Full source on GitHub ↗

Its description: Install, open or connect ChatCut Desktop in Antigravity (安装 ChatCut), or start editing or recreating a video with it: edit or cut a video, add captions or subtitles, clean up a talking-head recording, add B-roll, music…

---
name: connect-chatcut-desktop
description: "Install, open or connect ChatCut Desktop in Antigravity (安装 ChatCut), or start editing or recreating a video with it: edit or cut a video, add captions or subtitles, clean up a talking-head recording, add B-roll, music or motion graphics, or recreate a viral or reference video with the user's own footage (剪视频、加字幕、剪口播、复刻爆款、复刻视频、做同款). Reuses an existing chatcut_desktop connection and Desktop's own Skills."
---

# Connect ChatCut Desktop

This plugin uses ChatCut Desktop only: a signed app on the user's computer that provides the local `chatcut_desktop` MCP server and its own Skills. It does not connect ChatCut Web. Existing Web projects stay in the ChatCut Web editor; Desktop does not migrate them.

A request to install ChatCut covers every step below. Run them in one go without stopping to ask for confirmation, then hand the user over to a new conversation (step 5), where the Desktop tools load.

## 1. Check the current connection

If `chatcut_desktop` tools are available in this conversation, call `get_active_project`. A result with `surface: "desktop"` means Desktop is installed and connected, even when `projectId` is null. Make sure Desktop's Skills are visible (step 4), then continue the user's request without reinstalling. If it reports `signedIn: false`, local work still runs; ask the user to sign in in the ChatCut window when an account action needs it.

## 2. Install Desktop on this computer

Confirm that commands run on the user's own macOS 13+ or Windows 10+ x64 computer with a graphical desktop. A Linux, headless or remote shell cannot install the app on another computer: explain that and keep the request for a supported computer.

Read the complete official guide with `curl -fsSL https://chatcut.io/desktop` (`curl.exe` on Windows). Follow it for installation and sign-in, and run its installer with this plugin's install source:

macOS:

```sh
curl -fsSL https://chatcut.io/install.sh | CHATCUT_INSTALL_SOURCE=antigravity_marketplace sh
```

Windows PowerShell:

```powershell
$env:CHATCUT_INSTALL_SOURCE='antigravity_marketplace'; irm https://chatcut.io/install.ps1 | iex
```

The download is about 740 MB; tell the user before it starts. Wait for the command to finish. Do not add sudo, disable signature or trust checks, force-quit ChatCut, or automate the Desktop UI. Before updating a running app, have the user save and quit it normally. A failed install is not success: report its actual error.

The installer does not register Antigravity. Lines about agent MCP setup failing, or a list of other agents, do not decide Antigravity's connection: continue with step 3.

The installer opens ChatCut and also a browser sign-in page. They are the same sign-in: tell the user to sign in once, in either the browser page or the ChatCut window. Do not wait for sign-in before continuing; setup does not need it.

## 3. Register `chatcut_desktop` in Antigravity

Until ChatCut Desktop registers Antigravity itself, do this right after installing:

1. Check that Desktop's launcher exists: macOS `$HOME/Library/Application Support/ChatCut/chatcut-mcp`, Windows `%APPDATA%\ChatCut\chatcut-mcp.cmd`. If it is missing, Desktop has not finished its first launch: ask the user to open ChatCut, then check again. Never create or edit the launcher.
2. Run `agy mcp list`. If `chatcut_desktop` already points at that launcher, keep it. If it points at an old ChatCut launcher that no longer exists, replace it. If it points at anything else, ask before replacing it, because `agy mcp add` overwrites an entry with the same name.
3. Register it with the expanded absolute launcher path:

   ```sh
   agy mcp add --env CHATCUT_MCP_CLIENT=antigravity chatcut_desktop "$HOME/Library/Application Support/ChatCut/chatcut-mcp"
   ```

   On Windows use the expanded `%APPDATA%\ChatCut\chatcut-mcp.cmd` path. `agy mcp add` writes `~/.gemini/config/mcp_config.json` and keeps other servers. Without `agy`, back up that file and merge `{"mcpServers":{"chatcut_desktop":{"command":"<launcher path>","env":{"CHATCUT_MCP_CLIENT":"antigravity"}}}}` by hand, preserving every other entry. Treat an empty file as `{}`; stop and report other invalid JSON instead of overwriting it.

Do not add a ChatCut Web server.

## 4. Make Desktop's Skills visible

Desktop keeps its Skills in `~/.agents/skills/chatcut-*` and lists them in `~/.agents/skills/.chatcut-desktop-skills.json`. Antigravity reads global Skills from `~/.gemini/config/skills`, and does not follow a `skills.json` that points elsewhere. Link each Desktop-managed Skill there. Re-run this after Desktop updates; it only touches links that point into Desktop's folder.

macOS:

```sh
src="$HOME/.agents/skills"; dst="$HOME/.gemini/config/skills"; manifest="$src/.chatcut-desktop-skills.json"
[ -f "$manifest" ] || { echo "ChatCut Desktop has not synced its Skills yet; open ChatCut and retry."; exit 1; }
mkdir -p "$dst"
for dir in "$src"/chatcut-*/; do
  name=$(basename "$dir"); link="$dst/$name"
  [ -f "$dir/SKILL.md" ] && grep -q "\"$name\"" "$manifest" || continue
  if [ -L "$link" ]; then
    case "$(readlink "$link")" in "$src"/*) ;; *) echo "kept $link"; continue ;; esac
  elif [ -e "$link" ]; then echo "kept $link"; continue; fi
  ln -sfn "$src/$name" "$link"
done
for link in "$dst"/chatcut-*; do
  [ -L "$link" ] && [ ! -e "$link" ] && case "$(readlink "$link")" in "$src"/*) rm "$link" ;; esac
done
ls "$dst"
```

Windows PowerShell (junctions need no administrator rights):

```powershell
$src = Join-Path $HOME '.agents\skills'; $dst = Join-Path $HOME '.gemini\config\skills'
$manifest = Get-Content (Join-Path $src '.chatcut-desktop-skills.json') -Raw | ConvertFrom-Json
New-Item -ItemType Directory -Force $dst | Out-Null
foreach ($name in $manifest.entries.PSObject.Properties.Name) {
  $target = Join-Path $src $name; $link = Join-Path $dst $name
  if (-not (Test-Path (Join-Path $target 'SKILL.md'))) { continue }
  if (Test-Path $link) {
    $item = Ge

Excerpt, first 6,000 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 c8e8b04

claude plugin install chatcut@chatcut-inc
claude mcp add-json --scope user chatcut '{"type":"http","url":"https://api.chatcut.io/api/external-mcp/mcp","oauth_resource":"https:/

Commands as the repository documents them: shown, never run by us. Our verdict covers commit c8e8b0479de1 only.

Works in which agents?

What its own text says, not a test

AgentStatusMore
Claude CodeMentioned in its text, not testedSkills for Claude Code · About Claude Code
Claude DesktopMentioned in its text, not testedSkills for Claude Desktop · About Claude Desktop
CodexMentioned in its text, not testedSkills for Codex · About Codex
CursorMentioned in its text, not testedSkills for Cursor · About Cursor
Gemini CLIMentioned in its text, not testedSkills for Gemini CLI · About Gemini CLI
“Mentioned” means the skill’s text names the agent. We have not run it there.

Audit history

Every audit, every source change

  1. ✕BLOCK D 66 / 100

    first audit · source c8e8b0479de1

README badge

For maintainers · shows the latest audit

ModelsAtlas: BLOCK DMODELSATLASBLOCK · D
[![ModelsAtlas: BLOCK D](https://modelsatlas.com/skills/chatcut-inc/agent-plugin/badge.svg)](https://modelsatlas.com/skills/chatcut-inc/agent-plugin/)

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

Similar skills

Add this badge to your README

[![ModelsAtlas: BLOCK D](https://modelsatlas.com/skills/chatcut-inc/agent-plugin/badge.svg)](https://modelsatlas.com/skills/chatcut-inc/agent-plugin/)

Questions about agent-plugin

What does the agent-plugin skill do?
AI video editor for agents. ChatCut plugin and MCP server for Codex, ChatGPT, Claude Code, Cursor, Grok, Gemini CLI, WorkBuddy, Doubao Work and Qwen Work: edit your own footage by chatting, add captions and subtitles, remove silences and filler words, create motion graphics, and export an editable t
Is agent-plugin safe to install?
No — not without reading the findings first. The audit graded it D (66/100) and found 10 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What can agent-plugin access on my machine?
The audit observed that it reaches the network and runs shell commands. Each of those is consistent with what it says it does. Secrets in the source: none found.
Which assistants does agent-plugin work with?
Its documentation mentions claude-code, claude-desktop, codex, cursor and gemini-cli. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (c8e8b0479de1), read on 2026-11-12. 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.