Back to skill

Security audit

foolproof-kitchen

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent meal-planning purpose, but it collects and saves sensitive health data while producing an HTML report with avoidable script and medical-safety risks.

Review carefully before installing. Use it only if you are comfortable sharing body metrics and health conditions and having a generated report saved in the workspace. Treat outputs as general nutrition guidance, not medical advice, especially for pregnancy, diabetes, hypertension, gout, allergies, kidney/liver disease, medications, or eating-disorder history. Prefer a version that adds privacy notice and consent, avoids raw or undercooked egg recommendations, bundles or integrity-pins scripts, and renders report data safely without innerHTML.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
assets/recipe_template.html:473
Finding

Client-Side Script Injection Through Unescaped Template Data and Unsafe HTML Rendering

Content
View full analysis
`
${item.label} ${item.value}
` ).join(''); } ``` ```javascript function renderMeals() { const container = document.getElementById('mealsContainer'); container.innerHTML = meals.map(meal => `
...[truncated 4090 chars]
Remediation
View remediation
` element. - Read it with `textContent` and parse it using `JSON.parse`. - Alternatively, provide data in a separate local JSON file. 3. **Eliminate unsafe `innerHTML` sinks** - Create elements with `document.createElement`. - Assign all user-derived values through `textContent`. - Set numeric style values only after strict numeric parsing and range validation. - Use `classList` with an allowlist rather than inserting arbitrary values into class attributes. 4. **Sanitize any intentionally supported markup** - If rich HTML is required, process it with a maintained allowlist-based sanitizer. - Disallow scripts, event-handler attributes, dangerous URL schemes, embedded documents, SVG, and MathML unless explicitly required. 5. **Validate generated data** - Enforce schemas for meals, dishes, ingredients, and shopping items. - Apply type checks, length limits, numeric bounds, and allowlists before report generation. 6. **Add defense-in-depth controls** - Apply a restrictive Content Security Policy that blocks inline scripts and limits script and connection sources. - Avoid granting the generated page access to privileged APIs or sensitive same-origin application data. ]]>

T08 · Insecure Dependencies

Note
Location
assets/recipe_template.html:7
Finding

Remote Executable Dependency Loaded Without Subresource Integrity

Content
View full analysis
``` ### Technical Analysis Every generated report loads and executes Chart.js from a third-party CDN when opened. Although the package version is pinned to `4.4.0`, the script element has no Subresource Integrity hash. The browser therefore cannot verify that the received file exactly matches a reviewed artifact. The template also lacks a restrictive Content Security Policy. Consequently, compromise of the CDN, package distribution path, or served artifact could result in modified JavaScript executing with the same browser privileges as the report. The dependency is not proven malicious. The issue is the absence of integrity enforcement and the unnecessary runtime reliance on a remote executable resource. ### Attack Path 1. An attacker compromises or interferes with the CDN, package publication, or dependency-delivery path. 2. The user opens a generated recipe report while the remote resource is reachable. 3. The browser requests the Chart.js file from the third-party CDN. 4. Because no integrity hash is specified, the browser accepts any successful JavaScript response from that URL. 5. The modified script executes within the generated report's browser origin. ### Impact Assessment A compromised dependency could: - Read and alter all information displayed in the recipe report, including body metrics and special dietary conditions. - Change nutritional or cooking instructions. - Display deceptive content. - Make attacker-controlled network requests. - Access other same-origin resources where browser policy permits. The script executes in the browser context only. The reviewed code does not demonstrate operating-system privileges, persistence, or direct ser ...[truncated 28 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (12)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 88)May include surrounding context.

md
使用 `assets/recipe_template.html` 作为基础模板,生成完整的 HTML 页面:

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 123)May include surrounding context.

md
使用 `assets/recipe_template.html` 作为基础模板,生成完整的 HTML 页面:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly requests sensitive health/body data such as age, sex, height, weight, pregnancy status, diabetes, hypertension, allergies, and other medical conditions, but provides no user-facing privacy notice, minimization guidance, or retention/disclosure statement. Because this data is highly sensitive, users may overshare without understanding how it will be processed or stored, increasing privacy and compliance risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The description centers the skill on Chinese dietary guidelines and presents the skill primarily in Chinese, but does not indicate that users may opt for another language or locale basis. Under the policy, forcing a specific language or locale without user opt-in is a natural-language policy violation unless the regional constraint is clearly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The HTML sets lang="zh-CN" and the page title/content are exclusively in Chinese, which imposes a specific language/locale on all users. The file does not indicate that this is an optional localization choice or that the skill is intentionally limited to a China-specific context.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The template loads Chart.js from a third-party CDN at runtime without any integrity pinning or local bundling. If the CDN response is tampered with, blocked, or replaced, arbitrary JavaScript would execute in the page context and could access the user's highly sensitive health/profile data rendered into the document.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This file gives individualized health-related guidance, formulas, and condition-specific recommendations for diabetes, hypertension, gout, pregnancy, fat loss, and muscle gain, but it does not clearly frame the content as general educational information or require medical escalation for higher-risk cases. In the context of a skill that uses personal body metrics to generate meal plans, users may rely on this as personalized medical nutrition advice, creating real safety risk from inappropriate calorie targets or contraindicated recommendations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

This recipe recommends blending raw egg whites into a shake with no safety guidance. Raw egg consumption carries salmonella risk, and the danger is heightened because the skill targets health-focused users and explicitly mentions special dietary populations, who may incorrectly assume the recipe is safe and medically appropriate.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The workflow instructs the agent to save generated HTML to the workspace outputs directory and present the file, but does not warn the user that the file may contain sensitive personal health metrics and nutritional inferences. Persisting such content to disk can expand exposure beyond the immediate chat context through logs, shared workspaces, backups, or later access by other tools/users.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

SQP-3 applies to all file types and covers natural-language policy violations such as forcing a specific language without user opt-in. This document presents all instructions and content only in Chinese, with no indication that the user can choose another language or that the locale restriction is intentional and documented.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The breakfast recipe explicitly offers a 4-minute soft/undercooked egg option without any food-safety warning. In a cooking skill aimed at novices and special-diet users, this can increase risk of foodborne illness, especially for pregnant, elderly, immunocompromised, or otherwise vulnerable users.

Content

No source excerpt is available for this finding.