Back to skill

Security audit

Video Deep Reader · 视频深读

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious with medium confidence because it reports an unauthenticated token-request path that can run attacker-controlled JavaScript, alongside concerns about browser-cookie access, setup dependencies, and a diagnostic endpoint.

Consider the suspicious verdict and the incomplete scan coverage before installing. In particular, weigh the reported code-execution issue and whether you are comfortable with browser-cookie access and the additional setup components. The scan does not establish malicious intent or prove that credentials were stolen, but partial analysis means it cannot establish that the skill is safe.

Vulnerability Patterns
  • 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
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
vendor/bgutil/server/src/session_manager.ts
Finding

Unauthenticated token requests can execute attacker-controlled JavaScript

Content
View full analysis

Impact

Arbitrary JavaScript execution with the Node.js server's permissions, including access to readable files and environment variables and the ability to modify files or launch processes. Container deployment limits this impact to the container and its accessible resources unless additional host access is configured.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
Findings (53)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The caption retrieval, private cache, timestamp lookup, and diagnostics support the stated purpose. However, cookie_options explicitly enables yt-dlp to read Chrome cookies or a cookie export and use them for requests. It also configures an optional JavaScript token provider. These are material credential-access and authenticated-access capabilities not disclosed by the public-caption description or declared permissions. They require explicit invocation; the code does not show automatic cookie collection or unrelated credential exfiltration. The absence of note generation in this helper chunk is not itself treated as a mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Fetching captions supports the declared purpose, but accessing browser authentication state is a security-relevant capability beyond the described public-caption or supplied-transcript workflow. The script implementation is not shown, so the chunk does not establish cookie extraction details, exfiltration, or successful access to restricted content.

Content

No source excerpt is available for this finding.

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · SKILL.md (reported line 122)May include surrounding context.

Cookie mode (bot check fix)

Shared/data-center egress IPs trigger YouTube's bot check ("Sign in to confirm you're not a bot") on the player/caption endpoints, while playlist listing still works. When a fetch fails that way, retry in cookie mode — only after the user has explicitly agreed to use their YouTube login:

bash
python3 <skill_dir>/scripts/fetch_video.py "<youtube-url>" --cookies chrome
# or with an exported cookies.txt:
python3 <skill_dir>/scripts/fetch_video.py "<youtube-url>" --cookies /path/to/cookies.txt

One-time setup (already done on this machine):

bash
.venv/bin/pip install yt-dlp-ejs bgutil-ytdlp-pot-provider
# bgutil POT provider script (generates the PO Token YouTube requires):
git clone --depth 1 https://github.com/Brainicism/bgutil-ytdlp-pot-provider.git vendor/bgutil
cd vendor/bgutil/server && npm install && npx tsc   # produces build/generate_once.js

Requirements and behavior:

  • --cookies chrome reads Chrome cookies directly; **C

Code capability 'env' detected in scripts/fetch_video.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'env' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · scripts/fetch_video.py (reported line 343)May include surrounding context.

python
xt": text}
            for seconds, stamp, text in selected
        ],
    }


def safe_error(error):
    """Return a bounded error without credential or bypass instructions."""
    value = str(error).lower()
    if "sign in" in value or "not a bot" in value or "cookies" in value:
        return (
            "YouTube requested sign-in or bot verification. Retry with cookie "
            "mode (--cookies chrome, requires Chrome quit + logged into YouTube), "
            "or paste/upload a transcript, or try later."
        )
    if "age" in value or "restricted" in value or "private video" in value:
        return (
            "The video is access-restricted. This skill does not bypass access controls; "
            "provide an authorized transcript instead."
        )
    if "unavailable" in value or "not available" in value:
        return "The video is unavailable. Check the link or provide a transcript."
    return "YouTube metadata or subtitle retrieval failed. Try later or provi

Cookie mode expands public-caption retrieval into browser credential access and authenticated fetching.

High
Category
Excessive Agency
Confidence
98% confidence
Finding

The Chrome option delegates browser-cookie extraction to yt-dlp without restricting extraction to YouTube cookies or selecting an isolated profile. This brings unrelated browser sessions into the credential-handling process, increasing the consequences of a compromised dependency beyond the authentication needed for caption retrieval; it is not evidence of intentional exfiltration.

Content

No source excerpt is available for this finding.

Privileged Container / Container Escape

High
Category
Privilege Escalation
Confidence
97% confidence
Finding

Host networking is not itself a privileged-container escape, but the recommendation creates a concrete exposure risk here: the documented image binds its unauthenticated server to all interfaces. On platforms supporting host networking, adding this option bypasses the loopback-only port publication and can expose the service directly on host interfaces while also expanding access to host-local services.

Content

Scanner excerpt · vendor/bgutil/README.md (reported line 68)May include surrounding context.

md
Our Docker image comes in two flavors: Node.js or Deno. The `:latest` tag defaults to Node.js, but you can specify an alternate version/flavor like so: `brainicism/bgutil-ytdlp-pot-provider:2.0.2-deno`. The `:node` tag also points to the latest Node.js image, and `:deno` points to the latest Deno image.

> [!IMPORTANT]
> Note that the docker container's network is isolated from your local network by default. If you are using a local proxy server, it will not be accessible from within the container unless you pass `--net=host` as well.

**Native:**

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · vendor/bgutil/plugin/yt_dlp_plugins/extractor/getpot_bgutil_script.py (reported line 109)May include surrounding context.

python
return ()

    def _jsrt_envs(self) -> dict:
        return os.environ.copy()

    def _jsrt_path_impl(self) -> str | None:
        jsrt_path = _determine_runtime_path(

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · vendor/bgutil/plugin/yt_dlp_plugins/extractor/getpot_bgutil_script.py (reported line 317)May include surrounding context.

python
)

    def _jsrt_envs(self) -> dict:
        process_env = os.environ.copy()
        process_env['DENO_NO_PROMPT'] = '1'
        process_env['DENO_NO_UPDATE_CHECK'] = '1'
        process_env['FORCE_COLOR'] = 'false'

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

This is not an MCP server reference, but the cookie-mode setup does introduce an actual supply-chain risk: it clones an unpinned repository and installs dependencies before running build tooling. Following these instructions can execute changed or compromised upstream code with the user's privileges; explicit installation approval does not establish code integrity.

Content

No source excerpt is available for this finding.

The guarantee that fetching never invokes an external executable contradicts the cookie-mode runtime.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The unconditional statement that fetching never invokes external executables conflicts with the documented Node-based solver and token-provider runtime. That inaccurate security guarantee can cause users or agents to approve credential-bearing execution without accounting for the additional executable and dependency trust boundaries.

Content

No source excerpt is available for this finding.

The error sanitizer promises no credential instructions but recommends using logged-in browser cookies.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L338 says errors are returned without credential or bypass instructions. The immediately following branch instead tells callers to retry with '--cookies chrome' and a browser logged into YouTube, actively recommending a credential-bearing recovery path.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx eslint'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx eslint'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx prettier'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx prettier'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Docker image referenced without tag or digest: 'docker run --name'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The complete Docker command uses an image without a tag or digest, implicitly selecting the mutable latest tag. Future pulls can therefore execute different provider code without a reviewed configuration change, creating a supply-chain and reproducibility risk.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% 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.

Content

Scanner excerpt · vendor/bgutil/README.md (reported line 177)May include surrounding context.

For provider versions >=1.2.0, you may have issues while installing the canvas dependency on Termux. The Termux environment is missing a android_ndk_path and two packages by default. Run the following commands to setup the dependencies correctly.

shell
mkdir ~/.gyp && echo "{'variables':{'android_ndk_path':''}}" > ~/.gyp/include.gypi
pkg install libvips xorgproto

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Diagnostic HTTP endpoint exposes internal token-minter cache state.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The unauthenticated diagnostic endpoint returns internal minter-cache keys and logs the full cache, potentially disclosing proxy URLs containing credentials as well as token-related state in logs. Loopback binding limits remote access by default, but does not authenticate local callers, and wider deployment exposure increases the risk; browser-origin filtering is not authentication.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · vendor/bgutil/server/build/session_manager.js (reported line 248)May include surrounding context.

js
(await this.getChallengeFromHomepage(potCtx)) ?? challenge;
            if (!challenge) {
                this.logger.debug("Using challenge from /att/get (legacy fallback)");
                const attGetResponse = await potCtx.fetch("https://www.youtube.com/youtubei/v1/att/get?prettyPrint=false", {
                    method: "POST",
                    headers: {
                        ...getHeaders(),

Known Vulnerable Dependency: basic-ftp==6.2.1 — 2 advisory(ies): CVE-2026-126345 (basic-ftp: Uncaught TypeError in MLSD directory-listing parser via __proto__ fac); CVE-2026-126346 (basic-ftp: Uncaught exception crashes the Node.js process when a password (or re)

Medium
Category
Supply Chain
Confidence
80% confidence
Finding

The lockfile includes basic-ftp 6.2.1, matching the version identified by the supplied advisories; the reported uncaught exceptions could terminate the Node.js process and interrupt transcript fetching. It is an indirect runtime dependency through proxy-agent → pac-proxy-agent → get-uri, so exploitation would require reaching affected FTP operations with attacker-influenced input. The video-caption workflow does not inherently require FTP, making this less exposed than an FTP-focused service; neither vulnerable-path reachability nor the advisory details can be independently verified from this lockfile alone.

Content

No source excerpt is available for this finding.