Back to skill

Security audit

yaps-background-removal

Security checks for vulnerabilities and agentic risk

Overview

The skill is presented as image background removal, but its bundled runtime also exposes broader Yaps command passthrough plus unrelated transcription and meeting workflows.

Review this package before installing if you only wanted background removal. It uses the local Yaps app and account session as disclosed, but the bundled runtime can invoke broader Yaps workflows, including transcription and meeting processing, so install it only if you are comfortable with that wider local-file command surface.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (19)

Description-Behavior Mismatch

High
Confidence
99% confidence
Finding
The file implements transcription and subtitle workflows even though the skill is described as background removal. That scope mismatch is dangerous because it exposes materially different data-handling behavior, including processing of arbitrary local audio/video files and producing transcripts, which a user or host would not reasonably expect from this skill.

Description-Behavior Mismatch

High
Confidence
99% confidence
Finding
The meeting transcription path adds audio extraction from video and creation of meeting transcripts, capabilities unrelated to background removal. In the context of this skill, that broadens access to sensitive local media and derived text content, increasing the chance of covert collection or misuse of private conversations.

Description-Behavior Mismatch

High
Confidence
98% confidence
Finding
The CLI entrypoints allow generic Yaps argument passthrough and request-mode execution, which enables workflows outside the advertised image-background-removal scope. This is dangerous because the host may grant trust based on the manifest, while the runtime can invoke broader Yaps features on local files than users consented to.

Credential Access

High
Category
Privilege Escalation
Content
function accountStatusUnsafeDiagnosis(appVersion) {
  return {
    code: "account_status_unsafe",
    message: `Yaps ${appVersion || "before 2.3.124"} uses an older credential-based account check, so this plugin deliberately did not run it. Update Yaps to 2.3.124 or newer; the plugin will then reuse the desktop sign-in, trial, or Yaps Pro automatically. Do not approve a Keychain prompt or create a separate plugin account.`,
  };
}
Confidence
70% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
function accountStatusUnsafeDiagnosis(appVersion) {
  return {
    code: "account_status_unsafe",
    message: `Yaps ${appVersion || "before 2.3.124"} uses an older credential-based account check, so this plugin deliberately did not run it. Update Yaps to 2.3.124 or newer; the plugin will then reuse the desktop sign-in, trial, or Yaps Pro automatically. Do not approve a Keychain prompt or create a separate plugin account.`,
  };
}
Confidence
70% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
function accountStatusUnsafeDiagnosis(appVersion) {
  return {
    code: "account_status_unsafe",
    message: `Yaps ${appVersion || "before 2.3.124"} uses an older credential-based account check, so this plugin deliberately did not run it. Update Yaps to 2.3.124 or newer; the plugin will then reuse the desktop sign-in, trial, or Yaps Pro automatically. Do not approve a Keychain prompt or create a separate plugin account.`,
  };
}
Confidence
70% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
function accountStatusUnsafeDiagnosis(appVersion) {
  return {
    code: "account_status_unsafe",
    message: `Yaps ${appVersion || "before 2.3.124"} uses an older credential-based account check, so this plugin deliberately did not run it. Update Yaps to 2.3.124 or newer; the plugin will then reuse the desktop sign-in, trial, or Yaps Pro automatically. Do not approve a Keychain prompt or create a separate plugin account.`,
  };
}
Confidence
70% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
message: "Yaps is signed in, but its trial or Yaps Pro access is not active. Open Yaps to review the available trial or Yaps Pro options; the plugin will pick up the change automatically.",
    };
  }
  if (auth.status === "credential_unavailable" || auth.diagnosticCode === "keychain_unavailable") {
    return {
      code: "account_status_unsupported",
      message: "This installed Yaps helper uses an older credential-based account check. Update Yaps and retry. Do not approve a Keychain prompt or create a separate plugin account.",
Confidence
70% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Lp3

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding
Without declared permissions the skill's intent is opaque and cannot be validated.

Intent-Code Divergence

Medium
Confidence
90% confidence
Finding
The comment says the adapter writes no diagnostics, credentials, settings, or host permissions, implying a narrowly scoped adapter surface. However, the code immediately performs account/session resolution and applies resolved settings to commands, which contradicts the comment's framing of what the adapter handles.

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
The adapter spawns the Yaps CLI with user-controlled arguments after only generic string validation, making the subprocess capability much broader than necessary for a background-removal tool. Even with shell=false, this still enables unintended helper functionality if upstream routing allows arbitrary commands, increasing attack surface and privilege through an overly capable backend.

Session Persistence

Medium
Category
Rogue Agent
Content
export const MIN_SAFE_AUTH_STATUS_VERSION = "2.3.124";
const MAX_PROBE_BYTES = 64 * 1024;
const MAX_VERSION_BYTES = 4 * 1024;
const MAX_PLIST_BYTES = 64 * 1024;
const MAX_WINDOWS_SHIM_BYTES = 4 * 1024;
const RUNNING_YAPS_PROCESS_NAMES = new Set(["yaps.exe", "yaps_mcp.exe", "yaps_cli.exe"]);
// Fixed WMI filter: never interpolate a discovered path into this command.
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Session Persistence

Medium
Category
Rogue Agent
Content
export const MIN_SAFE_AUTH_STATUS_VERSION = "2.3.124";
const MAX_PROBE_BYTES = 64 * 1024;
const MAX_VERSION_BYTES = 4 * 1024;
const MAX_PLIST_BYTES = 64 * 1024;
const MAX_WINDOWS_SHIM_BYTES = 4 * 1024;
const RUNNING_YAPS_PROCESS_NAMES = new Set(["yaps.exe", "yaps_mcp.exe", "yaps_cli.exe"]);
// Fixed WMI filter: never interpolate a discovered path into this command.
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Session Persistence

Medium
Category
Rogue Agent
Content
if (platform === "darwin") {
    const application = macApplicationForCli(cli.path, canonicalize);
    if (!application) return null;
    const plist = posix.join(application, "Contents", "Info.plist");
    try {
      const metadata = statFile(plist);
      if (!metadata.isFile() || metadata.size > MAX_PLIST_BYTES) return null;
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Session Persistence

Medium
Category
Rogue Agent
Content
if (platform === "darwin") {
    const application = macApplicationForCli(cli.path, canonicalize);
    if (!application) return null;
    const plist = posix.join(application, "Contents", "Info.plist");
    try {
      const metadata = statFile(plist);
      if (!metadata.isFile() || metadata.size > MAX_PLIST_BYTES) return null;
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Session Persistence

Medium
Category
Rogue Agent
Content
if (platform === "darwin") {
    const application = macApplicationForCli(cli.path, canonicalize);
    if (!application) return null;
    const plist = posix.join(application, "Contents", "Info.plist");
    try {
      const metadata = statFile(plist);
      if (!metadata.isFile() || metadata.size > MAX_PLIST_BYTES) return null;
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Session Persistence

Medium
Category
Rogue Agent
Content
if (platform === "darwin") {
    const application = macApplicationForCli(cli.path, canonicalize);
    if (!application) return null;
    const plist = posix.join(application, "Contents", "Info.plist");
    try {
      const metadata = statFile(plist);
      if (!metadata.isFile() || metadata.size > MAX_PLIST_BYTES) return null;
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Session Persistence

Medium
Category
Rogue Agent
Content
if (platform === "darwin") {
    const application = macApplicationForCli(cli.path, canonicalize);
    if (!application) return null;
    const plist = posix.join(application, "Contents", "Info.plist");
    try {
      const metadata = statFile(plist);
      if (!metadata.isFile() || metadata.size > MAX_PLIST_BYTES) return null;
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Session Persistence

Medium
Category
Rogue Agent
Content
if (platform === "darwin") {
    const application = macApplicationForCli(cli.path, canonicalize);
    if (!application) return null;
    const plist = posix.join(application, "Contents", "Info.plist");
    try {
      const metadata = statFile(plist);
      if (!metadata.isFile() || metadata.size > MAX_PLIST_BYTES) return null;
Confidence
75% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Static analysis

Detected: suspicious.dangerous_exec

Shell command execution detected (child_process).

Critical
Code
suspicious.dangerous_exec
Location
runtime/run.mjs:74