Back to skill

Security audit

podcast-digest

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent Chinese-language podcast digest helper with no install-time or persistent behavior, but generated HTML reports should be handled carefully when they include untrusted podcast text.

Install this if you want Chinese-oriented podcast analysis and visual summaries. Use it for clear podcast or audio-content tasks, and treat generated HTML or Web PPT output as untrusted until podcast-derived fields are escaped or sanitized, especially before publishing it on an authenticated or shared website.

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

Warning
Location
references/web-report-template.md.md:247
Finding

Unescaped Podcast Content in Generated Web Reports

Content
View full analysis
EP[XX] · [播客名]

[本期标题]

[核心主题一句话描述]

``` The corresponding template instruction is: ```markdown 1. 将每个 `[占位符]` 替换为实际内容 ``` Translated: replace every placeholder with actual content. ### Technical Analysis The template places podcast-derived or user-controlled fields directly into HTML element content. Its filling instructions require direct placeholder replacement but do not require contextual HTML escaping or sanitization. Podcast titles, transcript excerpts, speaker names, quotes, summaries, and metadata may originate from user input or externally retrieved podcast information. If a value contains active HTML, replacing a placeholder verbatim turns that value into document markup rather than inert text. For example, a malicious title containing an element with an event handler could execute script when the generated report is opened. The template's own JavaScript is benign; the vulnerability arises from unsafe insertion of untrusted content into the generated HTML. ### Attack Path 1. An attacker supplies a crafted podcast title, transcript, quote, or metadata value containing active HTML. 2. The Skill analyzes that content and generates a Web report. 3. Following the template instruction, the Agent replaces an HTML placeholder with the crafted value without escaping it. 4. The resulting report preserves the attacker-controlled HTML as executable markup. 5. A user opens or hosts the generated report in a browser. 6. The injected markup or event handler executes in the security context assigned to the generated document. ### Impact Assessment Successful exploitation can m ...[truncated 531 chars]
Remediation
View remediation
`, `"`, and `'`. 2. Generate dynamic nodes through DOM APIs and assign untrusted text using `textContent` rather than `innerHTML`. 3. If limited formatting is intentionally supported, sanitize it with a maintained strict allowlist that removes: - `

T09 · Insecure Skill Coding Practices

Warning
Location
references/webppt-template.md:170
Finding

Unescaped Podcast Content in Generated Web Slide Decks

Content
View full analysis
🎙️ [播客名] · EP[XX]
[本期标题]
[副标题/核心问题]
``` ```html
  • [核心问题1]
  • [核心问题2]
  • [核心问题3]
``` ```html
[金句文本]
— [说话人] · EP[XX]
``` ### Technical Analysis The slide template embeds podcast titles, questions, quotations, and speaker names directly inside HTML elements. No escaping, sanitization, or safe rendering requirement is specified. These values may be copied from attacker-controlled transcripts or metadata. Verbatim replacement of the placeholders allows an attacker to terminate the intended textual context and insert arbitrary HTML elements or event-handler attributes. When the generated slide deck is opened, the browser interprets the injected value as markup. The built-in slide-navigation JavaScript does not itself process untrusted input. The exploitable condition is the template's unsafe content-generation model. ### Attack Path 1. An attacker places malicious HTML in a podcast title, transcript quotation, speaker name, or core question. 2. The malicious value is selected for inclusion in a generated Web PPT. 3. The Agent inserts the value directly into one of the HTML placeholders. 4. The generated slide deck contains active attacker-controlled markup. 5. A viewer opens or hosts the deck in a browser. 6. The injected content executes or manipulates the presentation in the document's browser security context. ### Impact Assessment An attacker may alter slide content, impersonate navigation or promp ...[truncated 489 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (5)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The natural-language description and operating instructions are entirely in Chinese and describe required behavior, but they do not indicate that output language should follow user preference or offer opt-in for Chinese. This may violate language/locale policy when used in a multilingual environment because it implicitly forces one language without user choice.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill’s activation rules are overly broad and mandatory, including generic requests like summarizing content, taking notes, and analyzing material. This can cause the skill to trigger in situations beyond podcasts, leading to incorrect tool routing, unintended network access for content lookup, or the application of podcast-specific processing to unrelated user data.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The HTML template hard-codes lang="zh-CN", and the surrounding content is entirely Chinese-language, which signals a fixed locale requirement. The file does not offer a language/locale option or explain that the template is intended only for a China-specific or Chinese-only context, so this may violate language-choice policy.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The entire template library and usage guidance are written only in Chinese, with no indication that users may choose another language or that the skill is intentionally limited to a Chinese-language audience. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The document title, all instructional text, and the HTML example specify Chinese content, including <html lang="zh-CN">, which enforces a specific language/locale. There is no indication that users can choose another language or that the template is intentionally limited to a China-specific compliance or regional use case.

Content

No source excerpt is available for this finding.