Back to skill

Security audit

HandDraw Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent video-rendering purpose, but review is warranted because untrusted animation assets can read outside the asset folder and inject raw SVG/HTML into the renderer.

Use this only with project JSON and SVG assets you trust until asset paths are constrained to the asset directory and SVGs are sanitized. Avoid putting sensitive or confidential text in narration unless you are comfortable sending it to Edge TTS. Review the install/update scripts before running them because they copy into live agent skill directories and install dependencies.

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

Warning
Location
packages/renderer/src/render.ts:9
Finding

Asset Path Traversal and Unsafe SVG Injection

Content
View full analysis
{ const issues = validateProject(project); if (issues.length) throw new Error(issues.map((issue) => `${issue.path}: ${issue.message}`).join("\n")); const prepared = structuredClone(project) as PreparedProject; for (const scene of prepared.scenes) for (const object of scene.objects) if (object.kind === "svg" && object.asset) object.assetContent = await readFile(join(assetDirectory, object.asset), "utf8"); const entryPoint = join(dirname(fileURLToPath(import.meta.url)), "index.js"); const serveUrl = await bundle({ entryPoint }); const composition = await selectComposition({ serveUrl, id: "HandDrawAnimation", inputProps: { project: prepared } }); await renderMedia({ composition, serveUrl, codec: "h264", outputLocation, inputProps: { project: prepared } }); } ``` ```tsx // packages/renderer/src/Root.tsx:49 return <>
{!narrative && }; ``` ### Technical Analysis The project validator only verifies that an SVG object has a non-empty `asset` property. It does not ensure that the value is a relative, normalized SVG path located beneath the configured asset directory. The renderer passes the untrusted asset value directly to: ```ts join(assetDirectory ...[truncated 2331 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
packages/audio/requirements.txt:1
Finding

Unpinned Python Dependency Installed During Hermes Setup

Content
View full analysis
=7.2.0 ``` ```bash # scripts/install-hermes.sh:15-17 if ! python3 -m edge_tts --version >/dev/null 2>&1; then python3 -m venv .venv .venv/bin/pip install -r packages/audio/requirements.txt fi ``` ### Technical Analysis The Python requirement specifies only a minimum version. Each installation can therefore resolve a different future version of `edge-tts` and different versions of its transitive dependencies. The Hermes installer automatically invokes `pip install` as part of the documented installation workflow. Although installation occurs in an isolated virtual environment, package installation and later package execution still run with the operating-system permissions of the user performing the installation. The dependency set is not protected by exact version pins or package hashes. This is a supply-chain hardening issue rather than evidence that the currently named package is malicious. The risk arises because the effective code installed in the future may differ from the code reviewed with this project. ### Attack Path 1. A user runs the documented `scripts/install-hermes.sh` installer. 2. If `edge-tts` is not already available, the script creates a local Python virtual environment. 3. Pip resolves the newest package release satisfying `edge-tts>=7.2.0`, along with its current transitive dependency graph. 4. A compromised, malicious, or unexpectedly incompatible future release is downloaded and installed. 5. The installed code is subsequently executed when narration synthesis invokes `python -m edge_tts`. 6. That code runs with the filesystem and network privileges of the user operating the Skill. ### Impact Assessment A compromised depen ...[truncated 463 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (48)

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

If the associated code performs filesystem synchronization or copying into user home directories, that behavior is unrelated to the stated animation-generation purpose and materially increases risk. Undisclosed write or sync operations can overwrite user files, establish persistence-like footholds, or exfiltrate sensitive local content through secondary workflows.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

If the associated code performs filesystem synchronization or copying into user home directories, that behavior is unrelated to the stated animation-generation purpose and materially increases risk. Undisclosed write or sync operations can overwrite user files, establish persistence-like footholds, or exfiltrate sensitive local content through secondary workflows.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 16)May include surrounding context.

md
5. Run `node scripts/check-environment.mjs` once on a new machine. Fix every missing dependency.

Known Vulnerable Dependency: browserslist==4.28.6 — 2 advisory(ies): CVE-2026-73088 (Browserslist: Uncaught crash / prototype write via untrusted browserslist-stats.); CVE-2026-73089 (Browserslist: Unbounded memory growth (no cache eviction) via distinct query res)

High
Category
Supply Chain
Confidence
92% confidence
Finding

The lockfile includes browserslist 4.28.6, which is flagged for crashes/prototype write via untrusted browserslist-stats and unbounded memory growth via crafted distinct query results. If any part of the skill or its toolchain accepts attacker-controlled browserslist configuration or stats data, this can lead to denial of service and potentially unsafe object mutation behavior. Because this project bundles and renders content, build tooling may process project-controlled metadata, making this more relevant than a purely dormant dependency.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: fast-uri==3.1.3 — 6 advisory(ies): CVE-2026-75931 (fast-uri vulnerable to host confusion via skipped IDN canonicalization on scheme); CVE-2026-18446 (fast-uri vulnerable to host confusion via backslash authority introducer); CVE-2026-75975 (fast-uri vulnerable to server-side request forgery via malformed IPv6 normalizat) +3 more

High
Category
Supply Chain
Confidence
87% confidence
Finding

fast-uri 3.1.3 is reported vulnerable to multiple URI parsing issues including host confusion and SSRF-related malformed authority handling. In a system that may fetch remote media, assets, or configuration based on URLs, incorrect normalization or parsing can undermine hostname allowlists and redirect requests to unintended internal or sensitive destinations. Even though this file only shows the dependency, URI parser flaws are meaningful in a media/rendering skill where external asset references are plausible.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: nanoid==3.3.16 — 1 advisory(ies): CVE-2026-67213 (nanoid: custom generators can loop indefinitely when size is zero)

High
Category
Supply Chain
Confidence
80% confidence
Finding

nanoid 3.3.16 is flagged for a custom-generator infinite loop when size is zero. This is only exploitable if the application uses the affected custom generator API with attacker-influenced size values, which is not evident from the lockfile alone, so practical exposure may be limited. Nevertheless, if reachable, it can hang a process and cause denial of service.

Content

No source excerpt is available for this finding.

Unvalidated Output Injection

High
Category
Output Handling
Confidence
99% confidence
Finding

dangerouslySetInnerHTML is a direct sink for untrusted content and is one of the clearest React XSS/output-injection antipatterns. Because this skill's purpose is rendering deterministic animation from JSON, allowing arbitrary embedded markup is more dangerous than necessary and widens the attack surface beyond the stated functionality.

Content

Scanner excerpt · packages/renderer/src/Root.tsx (reported line 49)May include surrounding context.

tsx
const visibleText = characters.slice(0, Math.ceil(characters.length * writeProgress)).join("");
    return <div style={{ ...style, width: object.width ?? 1200, height: "auto", fontFamily: narrative ? "Kaiti SC, STKaiti, KaiTi, serif" : "PingFang SC, Microsoft YaHei, sans-serif", fontWeight: narrative ? 500 : 700, fontSize: object.height ?? 48, lineHeight: 1.48, letterSpacing: narrative ? "0.08em" : "0.045em", color: narrative ? "#1c1917" : "#172033", textShadow: narrative ? "0.4px 0.4px 0 #a8a29e" : "0.8px 0.8px 0 #dbe3eb", whiteSpace: "pre-wrap" }}>{visibleText}</div>;
  }
  return <><div style={style} className="handdraw-svg" dangerouslySetInnerHTML={{ __html: object.assetContent ?? "" }} />{!narrative && <DrawingHand object={object} progress={draw} />}</>;
}

function SceneLayer({ scene, time, opacity, narrative }: { scene: PreparedProject["scenes"][number]; time: number; opacity: number; narrative: boolean }) {

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
86% confidence
Finding

The skill instructs the agent to run local commands and perform environment checks, but it does not declare any explicit tool scope or permissions boundaries. In a skill that invokes CLI tooling and inspects dependencies, missing scope declarations can cause overbroad execution in host environments and makes it harder to enforce least privilege.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The default prompt 'Use $handdraw-skill to create a whiteboard animation video.' is broad and does not restrict when the skill should be invoked or what user intent must be present. This can cause over-triggering in loosely related conversations, leading the agent to route content into a code- or media-generating skill unexpectedly and potentially process sensitive or unsafe requests without sufficient gating.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The module executes ffprobe, ffmpeg, and a Python module via child_process to synthesize and process audio. For a skill described only as creating deterministic hand-drawn animations from a JSON DSL, arbitrary dependency on external executables and a Python TTS package is a broader capability than the stated purpose justifies.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The manifest describes creating deterministic hand-drawn/whiteboard MP4 animations from a JSON Animation DSL, but this code adds a separate capability: generating narration by invoking the Edge TTS service. While mixing local audio into a video is compatible with video creation, external speech synthesis is not stated in the manifest and materially expands what the skill does.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The function sends clip.text to the Edge TTS service without any visible consent, warning, or data-classification check. If users include sensitive, proprietary, or personal information in narration text, that content may be transmitted to a third-party service unexpectedly, creating a privacy and compliance risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The code defaults to zh-CN-XiaoxiaoNeural when no voice is specified, which imposes a specific language/locale choice on users automatically. There is no nearby indication of user choice, opt-in, or documented justification for this locale-specific default.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The CLI adds an external Edge TTS synthesis path when narration is present, which exceeds the stated deterministic local JSON-to-animation rendering behavior. This can transmit user-provided narration text to an external service and introduces undeclared network-dependent behavior, creating privacy, compliance, and trust risks for users who expect fully local rendering.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Network-backed speech synthesis is not obviously required by the declared hand-drawn animation purpose and is triggered from the CLI based on project content. In this context, the undeclared external dependency is more dangerous because users may supply sensitive script text, assume deterministic offline processing, and unknowingly cause data egress to a third party.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The component injects object.assetContent directly into the DOM with dangerouslySetInnerHTML and no sanitization. If animation JSON or embedded asset content is attacker-controlled, malicious SVG/HTML can trigger script execution, external resource loading, event-handler abuse, or renderer compromise depending on the execution environment.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Untrusted markup is inserted without any disclosure or trust boundary, so users and integrators may assume the renderer is deterministic while it actually processes active DOM content. In this skill context, that increases the chance that hostile animation assets are accepted and rendered unsafely.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file states that voice defaults to zh-CN-XiaoxiaoNeural and that narrative-sketch uses handwritten Chinese typography. This is a natural-language locale constraint presented as the default behavior, with no indication that users can choose another language or opt in to the Chinese-specific mode.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script automatically rsyncs the repository into two user skill-installation directories if they exist, with no prompt, dry-run, or explicit confirmation. This can silently overwrite local changes or propagate unintended/malicious repository content into agent-consumed skill paths, which is more sensitive here because the targets are active skill directories used by automation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The narration text and selected voice are fixed to Chinese (zh-CN-XiaoxiaoNeural) throughout the file, with no indication that the user can choose another language or locale. The policy requires flagging language or locale constraints when they are imposed without explicit user opt-in or documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The SVG contains the hard-coded label "糖分", which presents content only in Japanese/Chinese characters with no alternative language option or locale choice. This can violate language/locale policy when a skill asset is expected to be broadly usable across users without explicit regional scoping.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

This JSON file contains user-facing title, narration, and on-screen captions entirely in Chinese, along with a fixed zh-CN voice selection. Under the policy rule for natural-language violations, this is a locale/language constraint without any visible opt-in or alternative language choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

This SVG embeds the labels "产品", "客户", and "收入" directly in Chinese, which constitutes a fixed language choice in natural-language content. Because there is no indication that the skill is region-specific or that users can opt into this locale, it may violate the language/locale policy for broadly applicable skills.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The SVG contains the literal text "增长", which presents content in Chinese with no indication that language selection is optional or region-specific. This can violate a language/locale policy when the skill assets are expected to be locale-neutral or user-selectable.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The narration is hard-coded to use the zh-CN Chinese neural voice for all spoken output. This imposes a specific language/locale choice in the skill content without any visible opt-in, alternative locale handling, or stated region-specific justification.

Content

No source excerpt is available for this finding.