Back to skill

Security audit

Byted Podcast Gen

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to generate podcasts as described, but it also has under-disclosed credential-management and logging behavior that users should review before installing.

Install only if you trust the publisher and the Volcengine/Bytedance speech-service workflow. Prefer supplying a pre-provisioned MODEL_SPEECH_API_KEY, avoid setting ARK_SKILL_API_BASE unless it is a verified HTTPS service endpoint, and do not submit confidential documents or private URLs unless you are comfortable with their contents being sent to the external provider and potentially appearing in local or centralized logs.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/api_key.py:95
Finding

Bearer Token Sent to an Unvalidated Environment-Controlled Endpoint

Content
View full analysis
str: base = (base or "").strip() if not base: return "" parsed = urlparse(base) path = parsed.path or "/" if not path.endswith("/"): path = f"{path}/" query = dict(parse_qsl(parsed.query)) query["Action"] = action query["Version"] = version return urlunparse(parsed._replace(path=path, query=urlencode(query))) ``` ### Technical Analysis When `MODEL_SPEECH_API_KEY` is unavailable, the Skill obtains `ARK_SKILL_API_KEY` and `ARK_SKILL_API_BASE` from environment variables. It then attaches the Ark key as an `Authorization: Bearer` header to requests targeting the supplied base URL. The code accepts any URL scheme, hostname, port, and path. It neither requires HTTPS nor restricts the hostname to an approved credential-management service. Consequently, an attacker who can influence the process environment or deployment configur ...[truncated 1864 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/protocols/protocols.py:422
Finding

Complete User-Provided Podcast Content Exposed in INFO Logs

Content
View full analysis
str: """String representation""" if self.type in [MsgType.AudioOnlyServer, MsgType.AudioOnlyClient]: if self.flag in [MsgTypeFlagBits.PositiveSeq, MsgTypeFlagBits.NegativeSeq]: return f"MsgType: {self.type}, EventType:{self.event}, Sequence: {self.sequence}, PayloadSize: {len(self.payload)}" return f"MsgType: {self.type}, EventType:{self.event}, PayloadSize: {len(self.payload)}" elif self.type == MsgType.Error: return f"MsgType: {self.type}, EventType:{self.event}, ErrorCode: {self.error_code}, Payload: {self.payload.decode('utf-8', 'ignore')}" else: if self.flag in [MsgTypeFlagBits.PositiveSeq, MsgTypeFlagBits.NegativeSeq]: return f"MsgType: {self.type}, EventType:{self.event}, Sequence: {self.sequence}, Payload: {self.payload.decode('utf-8', 'ignore')}" return f"MsgType: {self.type}, EventType:{self.event}, Payload: {self.payload.decode('utf-8', 'ignore')}" ``` The session sender logs that complete message at INFO level: ```python async def start_session( websocket: websockets.WebSocketClientProtocol, payload: bytes, session_id: str ) -> None: """Start session""" msg = Message(type=MsgType.FullClientRequest, flag=MsgTypeFlagBits.WithEvent) msg.event = EventType.StartSession msg.session_id = session_id msg.payload = payload logger.info(f"Sending: {msg}") await websocket.send(msg.marshal()) ``` The logged payload is populated with user-controlled text and URLs in `scripts/podcast.py:83-105`: ```python req_params = { "input_id": args.input_id, "input_text": args.text, "nlp_texts": nlp_texts, ...[truncated 2428 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
Findings (16)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

A description-behavior mismatch is especially dangerous in agent skills because reviewers and users may approve the skill for podcast generation while it actually performs credential-management actions such as listing or creating API keys and persisting secrets to a .env file. Hidden secret-handling behavior can lead to credential leakage, unauthorized account changes, and trust-boundary violations, and the mismatch is made worse because the documented functionality does not justify those extra operations.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/api_key.py (reported line 64)May include surrounding context.

python
def _dotenv_path() -> str:
    return os.path.join(os.path.dirname(os.path.abspath(__file__)), ".env")


def _load_dotenv_if_available(dotenv_path: str) -> None:

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code uses one credential (ARK_SKILL_API_KEY) to list existing API keys or create a new speech API key, then stores the new secret for later use. That is unjustified credential provisioning inside a content-generation skill and creates a privilege-escalation path where compromise of one token can mint additional long-lived credentials.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill declares no explicit tool scope or permissions even though its instructions clearly require shell execution, environment variable access, file handling, and network access. This weakens containment and reviewability: an agent may invoke broader capabilities than a user would reasonably expect, increasing the chance of unsafe execution or data exposure.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The description is written as a Chinese-only instruction ('将某个话题或者网页内容总结合成为播客音频') and the skill documentation does not offer any language or locale choice for how the podcast should be generated. This creates a natural-language locale constraint without user opt-in or an explicit region-specific justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instructions encourage processing topics, URLs, downloaded documents, uploaded files, and long context through an external speech service but do not warn the user that this content will leave the local environment. This creates a real privacy and compliance risk, especially when files or long context may contain confidential, regulated, or personal data.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The activation criteria are broad enough that the skill may trigger on generic requests about podcasts, webpages, uploaded files, or long context, causing unintended execution. In this skill's context, accidental activation is more dangerous because it can send user-provided text, URLs, or file contents to external services and invoke shell/network operations without clear user intent.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill contains logic to enumerate and create API keys, then cache the resulting secret locally, which goes beyond simple podcast generation and expands the skill's privilege surface. Even if intended for convenience, automatic credential provisioning introduces unnecessary secret-handling and account-modifying behavior that could be abused or surprise users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The code sends HTTP POST requests to an external base URL using a Bearer authorization header and payload fields such as ProjectName and Name. There is no visible confirmation, logging, or explanatory comment/docstring informing users that external API calls with credentialed headers will be made.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The code persists MODEL_SPEECH_API_KEY into a local .env file without any user approval, warning, or lifecycle controls. Local secret persistence increases exposure to accidental disclosure through backups, packaging, repository mistakes, or access by other local processes/users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This code opens a WebSocket connection to an external endpoint and sends request parameters containing user-provided text, URLs, and speaker information. Although the operation is part of TTS generation, this file provides no confirmation prompt, warning comment, or visible user-facing notice that the supplied content will be sent off-host to a third-party service.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The parser help and argument set imply broader behavior than the code supports: --action help references action 0 and 4 while _generate still contains dead checks for actions 3 and 4, and --texts_output suggests podcast text will be written to a file. In reality, choices restrict --action to [0,4] and --texts_output is never used, so the documentation-like CLI interface actively misrepresents actual behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

This code performs network transmission over a websocket and includes session identifiers and arbitrary payload bytes in sent and received messages. While it logs activity, the logging is developer-facing rather than a user disclosure, and the file provides no user-facing warning, prompt, or comment indicating that user or system data will be transmitted externally.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
95% confidence
Finding

The dependency is specified as websockets>=12.0, which allows any future major or minor release to be installed. This weakens build reproducibility and can unintentionally introduce vulnerable or breaking versions through normal dependency resolution or supply-chain compromise. In this podcast-generation skill, the library is likely network-facing, so dependency drift is somewhat more security-relevant than for a purely local utility.

Content

Scanner excerpt · requirements.txt (reported line 1)May include surrounding context.

text
websockets>=12.0

Unverifiable Dependency: websockets has 4 known advisory(ies) (CVE-2018-1000518 (websockets is vulnerable to denial of service by memory exhaustion); CVE-2021-33880 (Observable Timing Discrepancy in aaugustin websockets library); CVE-2018-1000518 (aaugustin websockets version 4 contains a CWE-409: Improper Handling of Highly C) +1 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
87% confidence
Finding

Because websockets is unpinned, it is impossible to verify from this manifest whether the installed version includes fixes for known advisories affecting some releases of the package. This creates uncertainty around exposure to issues such as denial of service or side-channel behavior, especially since the skill appears to rely on websocket-based communication for podcast synthesis.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The script saves generated podcast audio to a local file path, either user-specified or a default output directory. While file output is expected functionality, there is no explicit warning or runtime notice here that the tool will create directories and write audio files on disk.

Content

No source excerpt is available for this finding.