Back to skill

Security audit

Media Sync

Security checks for vulnerabilities and agentic risk

Overview

The skill does what it claims, but its downloader accepts unvalidated URL input in a way that can turn a pasted link into command execution inside the skill runtime.

Review before installing. Use only in a constrained runtime with write access limited to the intended media directory, avoid pasting untrusted or unusual URL strings, and prefer a version that validates http/https URLs, passes -- before yt-dlp URLs, hardens path containment, and stores progress in a private state directory.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (4)

T09 · Insecure Skill Coding Practices

Error
Location
downloader.sh:35
Finding

User-Controlled yt-dlp Option Injection Enables Arbitrary Command Execution

Content
View full analysis
"$YTDLP_EXIT_FILE" ) 2>&1 | ``` ### Technical Analysis The `links` parameter is accepted as a space-separated string and split into individual shell array elements. Each resulting token is passed to `yt-dlp` as an argument without validating that it is an HTTP or HTTPS URL. Although `"$URL"` is shell-quoted, shell quoting does not prevent the invoked program from interpreting a value beginning with `-` as a command-line option. The script also does not place the standard `--` option terminator before the URL. Consequently, an attacker can provide `yt-dlp` options through the `links` parameter. This is particularly dangerous because `yt-dlp` supports execution-related options such as `--exec`. An injected execution option can cause commands to run under the operating-system identity and permissions assigned to the skill. ### Attack Path 1. An attacker supplies a crafted `links` value containing a token such as an injected `yt-dlp` execution option followed by a valid URL. 2. `read -ra LINKS` splits the string into separate arguments. 3. The injected token is assigned to `URL` during one loop iteration. 4. The script invokes `yt-dlp` without a `--` option terminator. 5. `yt-dlp` interprets the attacker-controlled token as an option instead o ...[truncated 888 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
downloader.sh:25
Finding

Destination Root Containment Can Be Bypassed Through Symbolic Links

Content
View full analysis
"$YTDLP_EXIT_FILE" ) 2>&1 | ``` ### Technical Analysis The scripts attempt to confine downloads under `/mnt/jellyfin_media` by: - Removing leading and trailing slashes. - Removing literal `..` sequences. - Prefixing the result with the configured root. This is not equivalent to canonical path validation. Filesystem operations such as `[[ -d ... ]]`, `mkdir -p`, and output-file creation follow symbolic links. If any component beneath `/mnt/jellyfin_media` is a symbolic link to a directory outside that root, the constructed path still appears to start with ` ...[truncated 1574 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
downloader.sh:11
Finding

Predictable Shared Temporary Files Permit Symlink Attacks and Status Tampering

Content
View full analysis
"$PROGRESS_LOG" < "$SNAP" <
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
manifest.yaml:4
Finding

Unpinned and Unverified Dependency Installation Creates Supply-Chain Risk

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (7)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The description claims the skill downloads media into /mnt/jellyfin_media subfolders and tracks progress. This code chunk only checks for a progress log at /tmp/media_sync_progress.log, parses session metadata and per-file snapshot logs in /tmp, and prints a status summary. It never initiates downloads, writes media files, or interacts with /mnt/jellyfin_media. While progress tracking is consistent with part of the description, the primary declared behavior of downloading media is absent in this code chunk, making the description inaccurate for the supplied code.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose says the skill downloads media into /mnt/jellyfin_media subfolders and tracks progress. This code does neither: it only inspects the filesystem under /mnt/jellyfin_media to validate or suggest folder names. Its primary purpose is folder discovery/validation, not media downloading. Access to the same root path is consistent, but the actual behavior is materially different from the declared description, so this is a mismatch.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · README.md (reported line 26)May include surrounding context.

brew install yt-dlp

Linux

sudo curl -L https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp -o /usr/local/bin/yt-dlp sudo chmod a+rx /usr/local/bin/yt-dlp

text

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · README.md (reported line 27)May include surrounding context.

brew install yt-dlp

Linux

sudo curl -L https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp -o /usr/local/bin/yt-dlp sudo chmod a+rx /usr/local/bin/yt-dlp

text

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill instructs an agent to download external media into /mnt/jellyfin_media, but it does not prominently warn that untrusted remote content will be written to the local filesystem. That omission can cause users to approve actions without understanding that network-fetched files will be stored persistently, increasing risk from unwanted content, legal/compliance issues, or disk consumption.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The manifest explicitly supports downloading arbitrary external URLs into a mounted media library, but it does not disclose the privacy, storage, bandwidth, and trust implications of fetching remote content and writing it directly into a persistent host-mounted path. In this context, the risk is elevated because the skill combines network retrieval with write access to /mnt/jellyfin_media, so a user may unknowingly import unwanted or harmful content, overwrite organizational structure, or consume significant system resources.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instructions require that all replies to the user "must be caveman-style," which imposes a mandatory communication style regardless of user preference. This is a natural-language policy concern because it removes normal user choice over language/locale or communication style without any opt-in or justification.

Content

No source excerpt is available for this finding.