Back to skill

Security audit

YouTube Daily Digest

Security checks for vulnerabilities and agentic risk

Overview

The skill largely does the advertised YouTube-to-Feishu digest job, but the published package includes apparent plaintext Feishu/OpenClaw credentials and uses local secrets with weak scoping guidance.

Review carefully before installing. Do not publish or install this package until the apparent secrets in scripts/config.json are removed and rotated. Use a virtual environment with pinned dependencies, provide the Feishu recipient explicitly, and confirm you are comfortable with YouTube transcript text being sent through the configured OpenClaw Gateway/Claude service and the resulting digest being posted to Feishu.

Vulnerability Patterns
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/config.json:20
Finding

Hardcoded Feishu and OpenClaw Credentials in Distributed Configuration

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:16
Finding

Unpinned Dependencies Installed into the System Python Environment

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/digest.py:204
Finding

Indirect Prompt Injection Through Untrusted Video Titles and Transcripts

Content
View full analysis
1 else result.strip() return title_zh, summary ``` ### Technical Analysis Video titles and transcripts are external, publisher-controlled content. The function interpolates them directly into the same prompt that contains the summarization instructions. It does not clearly isolate them as untrusted data, instruct the model to ignore instructions appearing inside them, or validate the semantic safety of the generated output. A publisher of a monitored channel, or an attacker who compromises such a channel, can place instruction-like text in a title or transcript. The model may interpret that text as an instruction rather than content to summarize. Output parsing only splits the response into a translated title and summary; it does not prevent malicious, misleading, or socially engineered text from being delivered in the Feishu card. The model call does not expose local tools to the transcript and the prompt does not include the Feishu secret or Gateway token. Therefore, the demonstrated impact is output manipulation rather than direct local code execution or established credential disclosure. ### Attack Path 1. An attacker controls or compromises one of the hardcoded YouTube channels. 2. The attacker publishes a recent vid ...[truncated 1095 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (15)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared purpose presents the skill as a simple YouTube digest, but the actual behavior includes reading local credentials, resolving recipient identity, calling a gateway/LLM service, sending data to Feishu, and persisting local state. This mismatch is dangerous because users may authorize or run the skill without understanding that sensitive local secrets and generated content will be accessed and transmitted to external services.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill sends transcript content to an external LLM endpoint through the configured gateway with no explicit notice, consent, or data-minimization controls. Transcript text can contain sensitive or proprietary information, and off-box transmission to a model provider materially changes the privacy and trust boundary.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill documentation exposes executable commands and describes behavior that requires broad capabilities (environment access, local file reads/writes, network, and shell execution) but declares no explicit tool scope or permission boundaries. That creates a real security gap because users and orchestration layers cannot constrain what the skill may access, especially since it reads local OpenClaw config and writes persistent state.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description says the AI generates Chinese summaries, and the architecture later specifies Chinese title translation and summaries. This imposes a language/locale behavior in natural-language instructions without documenting user choice, opt-in, or a justified region-specific requirement.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger phrases are generic enough that ordinary conversation about YouTube summaries or daily digests could invoke the skill unintentionally. Because this skill performs network activity, reads local configuration, and can send outbound messages, accidental invocation has meaningful side effects beyond a harmless lookup.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The markdown omits a clear warning that the skill reads local OpenClaw credentials/configuration and sends content through external services including an LLM/gateway and Feishu. This lack of transparency can lead users to expose secrets or sensitive summaries without informed consent, especially given the 'zero configuration' framing that downplays security implications.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The guide instructs the reader to read a local gateway authentication token from ~/.openclaw/openclaw.json and use it for API access, but it does so without any warning that this is a credential and without guidance on minimizing exposure. In a skill context, normalizing direct access to local secrets increases the chance that future automation reads, logs, or transmits the token unsafely, enabling unauthorized use of the user's local gateway.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions require the first line to be a Chinese title translation and the rest of the summary format is also specified in Chinese, which imposes a language choice. SQP-3 applies to all file types and flags language-policy violations when a specific language is forced without user opt-in or clear justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The script silently reads ~/.openclaw/openclaw.json to obtain Feishu app credentials and the gateway token, giving the skill access to sensitive secrets outside its own directory. Even if used for legitimate functionality, implicit secret harvesting increases blast radius if the skill is modified, compromised, or run in an untrusted context.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The manifest describes a daily YouTube digest that fetches selected channels, summarizes transcripts, and pushes Feishu cards. Spawning an external command-line tool via subprocess introduces a broader execution capability than the stated purpose requires, especially since transcript retrieval is already implemented through a Python API path.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/digest.py (reported line 158)May include surrounding context.

python
import tempfile, glob, re
    try:
        with tempfile.TemporaryDirectory() as tmpdir:
            subprocess.run([
                "yt-dlp", "--write-sub", "--write-auto-sub",
                "--sub-lang", "en", "--skip-download", "--sub-format", "vtt",
                "-o", f"{tmpdir}/sub",

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The system message explicitly instructs the model to produce Chinese summaries, imposing a fixed language choice. The policy requires flagging language or locale constraints when the skill forces a specific language without offering a user choice or documenting a justified region-specific requirement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The prompt requires the first line to be a Chinese translation and the remainder to be a Chinese summary, which enforces a specific output language. There is no evidence in this file of user opt-in or a documented reason that the skill must be Chinese-only.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

These functions call Feishu APIs using app_id, app_secret, bearer tokens, and message content, but the code does not provide a clear warning that external network requests will be made with local credentials and generated content. For code files, network transmission of user or system data should have some visible disclosure when not otherwise documented.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The skill enumerates local OpenClaw identity files and environment variables to discover a Feishu recipient ID without explicit user consent or tight scoping. That is broader-than-necessary access to local account metadata and can result in messages being sent to an unintended identity or expose internal identifiers to unreviewed skill code.

Content

No source excerpt is available for this finding.