Back to skill

Security audit

叙事手绘故事视频

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coherent narrated-video workflow, but one TTS helper can write plan-controlled audio outputs outside the intended project folder and its Python install path is not reproducibly pinned.

Review the TTS voice-plan file before running it, keep segment output paths relative under the renderer public directory, avoid --overwrite unless you intend to replace files, and consider installing the Python TTS stack in an isolated virtual environment or container with pinned dependency versions. The skill does not show malicious behavior, but its file-write scoping and mutable dependency installation deserve review before use.

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
scripts/synthesize_qwen3.py:72
Finding

Voice-plan path traversal allows files to be written outside the designated output directory

Content
View full analysis

Vulnerability Details

File Location: scripts/synthesize_qwen3.py, lines 72–88
Vulnerability Type: Unrestricted file path and arbitrary file overwrite
Risk Level: Medium

Vulnerable Code

python
for segment in plan["segments"]:
    output = args.public_dir / segment["output"]
    if output.exists() and not args.overwrite:
        print(f"Skip {output}")
        continue
    voice = plan["voices"][segment["speaker"]]
    instruction = "。".join(part for part in [voice.get("instruct", "").strip(), segment.get("instruct", "").strip()] if part)
    kwargs = {"text": segment["text"], "language": segment.get("language", language), "instruct": instruction or None}
    if mode == "voice-design":
        wavs, rate = model.generate_voice_design(**kwargs)
    else:
        speaker = voice.get("speaker")
        if not speaker:
            raise ValueError(f"{segment['speaker']}: custom-voice requires speaker")
        wavs, rate = model.generate_custom_voice(**kwargs, speaker=speaker)
    output.parent.mkdir(parents=True, exist_ok=True)
    sf.write(output, wavs[0], rate)

Technical Analysis

The output path is constructed by joining the user-selected public_dir with the output value taken directly from the JSON voice plan. The code does not reject:

  • Absolute paths, which cause pathlib to discard the intended base directory.
  • Parent-directory components such as ../.
  • Symbolic-link destinations.
  • Output paths outside the renderer's public asset directory.

Consequently, a voice plan can direct generated audio to any filesystem location writable by the process. Existing files are skipped by default, but they are overwritten when the caller supplies --overwrite. Files that do not already exist can be created outside public_dir without that option.

Validation in require_plan() only checks whether the output value is present; it does not validate or confine the path. ...[truncated 1435 chars]

Remediation
View remediation

Remediation Suggestions

  1. Reject absolute output paths and parent-directory traversal before performing any writes.

  2. Resolve both the base directory and candidate destination, then verify that the destination remains under the base:

    python
    public_dir = args.public_dir.expanduser().resolve()
    relative_output = Path(segment["output"])
    
    if relative_output.is_absolute() or ".." in relative_output.parts:
        raise ValueError("segment output must be a relative path without parent traversal")
    
    output = (public_dir / relative_output).resolve()
    if not output.is_relative_to(public_dir):
        raise ValueError("segment output escapes public_dir")
    
  3. Restrict output extensions to expected audio formats such as .wav.

  4. Refuse symbolic-link destinations and consider using secure, exclusive file creation.

  5. Require per-file confirmation before replacing existing files rather than relying only on a global --overwrite option.

  6. Apply the same path-confinement validation to all plan-controlled file paths.

T08 · Insecure Dependencies

Warning
Location
requirements-qwen3.txt:1
Finding

Unpinned and unhashed Python dependencies create a mutable supply-chain boundary

Content
View full analysis

Vulnerability Details

File Location: requirements-qwen3.txt, lines 1–4; installation command in README.md, lines 40–42
Vulnerability Type: Non-reproducible dependency installation
Risk Level: Medium

Vulnerable Code

requirements-qwen3.txt:

text
qwen-tts>=0.1.1
torch>=2.5
soundfile>=0.12
huggingface-hub>=0.25

Documented installation command:

bash
python3 -m venv ../.venv
../.venv/bin/pip install -r ../requirements-qwen3.txt

Technical Analysis

Every Python dependency uses an open-ended minimum-version constraint. No exact versions, artifact hashes, lockfile, or upper bounds are provided. The code installed by the documented command can therefore change over time without any modification to the audited repository.

Python packages can execute code during source-distribution build steps and whenever installed modules are imported. This is especially significant here because the Skill imports and runs large machine-learning packages and model-management libraries.

The weakness does not prove that any currently named package is malicious. It means that installation is not reproducible and does not ensure that users receive the versions reviewed by the project maintainers.

Attack Path

  1. A user follows the installation procedure in README.md.
  2. pip resolves the newest package versions satisfying the open-ended constraints.
  3. A future compromised release, maliciously replaced artifact, or otherwise unsafe version satisfies one of those constraints.
  4. The package is downloaded and installed because no exact version or expected artifact hash is enforced.
  5. Malicious code can execute during a source build or later when the Skill imports the affected package.
  6. That code runs with the privileges and environment access of the user performing installation or invoking the TTS script.

Impact Assessment

A compromised dependency can obtain a ...[truncated 450 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace minimum-version constraints with exact versions that have been tested and reviewed.

  2. Generate a hash-locked requirements file containing hashes for every permitted distribution.

  3. Install dependencies with hash verification:

    bash
    python3 -m pip install --require-hashes -r requirements-qwen3.lock
    
  4. Maintain separate lockfiles for supported Python versions and platforms where binary artifacts differ.

  5. Use a trusted package index explicitly and prevent unintended fallback to additional indexes.

  6. Add automated dependency vulnerability and provenance scanning.

  7. Review and deliberately update locked versions rather than accepting new releases automatically.

  8. Consider installing the TTS stack in a container or similarly isolated environment with minimal filesystem and credential access.

Vulnerability Patterns
  • 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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (50)

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The code is broadly related to story-video generation and does support text or image driven planning/import/render workflows for hand-drawn animations, so the high-level domain matches. However, the declared description emphasizes a richer multimedia pipeline with sentence-synchronous Chinese narration, optional character voices, subtitles, and background music. This code chunk is only a convenience entry point that shells out to npm scripts and explicitly reports rendering silent videos. There is no visible handling of audio synthesis, subtitle production, or BGM. Because those are central parts of the declared purpose rather than incidental implementation details, the description overstates what this supplied code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description promises a full Chinese story short-video creation pipeline with visuals, narration, subtitles, and music. The supplied code chunk is only a minimal application bootstrap for a Remotion project. While this entry point could belong to a larger video-rendering system, the code shown does not demonstrate the declared capabilities or primary purpose beyond generic renderer initialization. Therefore, based on the supplied chunk alone, the description materially overstates what the code actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description presents a comprehensive Chinese narrated short-video creation capability, including visual generation, narration, subtitles, and music mixing. The supplied code chunk does not implement any of those functions. It merely loads an uploaded storyboard from JSON and calculates total frames, which is at most a small supporting utility for a larger video workflow. Because the actual behavior shown is far narrower and materially different from the declared primary purpose, this is a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The description claims a broad multimedia pipeline for Chinese story short videos, including visuals, poster, subtitles, synchronized narration, and BGM mixing. The actual code only performs one narrow part of that pipeline: local Qwen3-TTS synthesis from a JSON plan into audio files. It does not generate or assemble video, create images, produce subtitles, or mix music. While TTS narration is consistent with part of the description, the code chunk's primary behavior is much narrower than the declared end-to-end short-video capability, so this is a material description/behavior mismatch.

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 · renderer/.gitignore (reported line 10)May include surrounding context.

text
codex-image-jobs.json
storyboard.generated.json
uploaded-pages.json
.env
.env.*
*.log
*.tgz

Known Vulnerable Dependency: browserslist==4.28.5 — 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

browserslist 4.28.5 is present and reportedly vulnerable to crash/prototype-write and unbounded memory growth when processing untrusted browserslist-stats or distinct attacker-controlled queries. In this skill, browserslist is a transitive tooling dependency rather than app runtime logic, which reduces exposure, but it can still affect CI/build environments or any feature that ingests external config/stat files.

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
88% confidence
Finding

fast-uri 3.1.3 is present and flagged for multiple URL parsing issues including host confusion and possible SSRF-adjacent behaviors. Here it is pulled in transitively through schema/validation tooling, so there is no evidence of direct network-facing use in this file, but vulnerable URL parsing code in the dependency graph is still a real supply-chain risk.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: nanoid==3.3.15 — 2 advisory(ies): CVE-2026-67214 (nanoid: non-secure generators can loop indefinitely with negative size); CVE-2026-67213 (nanoid: custom generators can loop indefinitely when size is zero)

High
Category
Supply Chain
Confidence
80% confidence
Finding

nanoid 3.3.15 is present and the cited issues describe infinite-loop/availability problems in non-secure or custom generator edge cases. This is a genuine vulnerable version, but the practical risk depends on whether the application or its dependencies invoke nanoid with attacker-influenced size parameters; in a video-rendering toolchain that risk appears limited.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: postcss==8.5.16 — 2 advisory(ies): CVE-2026-69153 (PostCSS: incomplete fix of GHSA-6g55-p6wh-862q — attacker-controlled sourceMappi); CVE-2026-73646 (PostCSS: Path Traversal in Previous Source Map Auto-Loading (sourceMappingURL) l)

High
Category
Supply Chain
Confidence
90% confidence
Finding

postcss 8.5.16 is present and flagged for source map handling issues including path traversal through sourceMappingURL processing. Because PostCSS commonly processes project-supplied CSS during builds, this can become more relevant if the skill accepts or renders untrusted assets/styles, potentially exposing local files or crashing build pipelines.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description states the pipeline is specifically for 'Chinese idiom, history, and children's-story shorts,' which imposes a language/locale constraint in the skill's natural-language behavior. The file does not offer users a language choice or explain why the locale restriction is required, so it fits the language/locale policy-violation category.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill instructs use of shell commands, environment variables, and local file paths, but it declares no explicit tool scope or permissions boundary. That creates an avoidable least-privilege gap: an agent may grant or assume broader execution and filesystem access than users expect, increasing the blast radius if the skill is misused or prompt-injected.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The manifest description says to use the skill for 'any Chinese narrated hand-drawn short-video request,' which is a broad natural-language trigger that could overlap with many ordinary requests for Chinese video creation. It does not provide explicit trigger boundaries, exclusions, or negative examples to clarify when this skill should or should not activate.

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 requires creation of 'Chinese story short videos' and the body repeatedly mandates Chinese title text and Chinese narration behavior, but does not indicate that the user can choose another language or explicitly opt into Chinese. This is a natural-language locale constraint that is forced rather than presented as a user-selectable option.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 55)May include surrounding context.

md
- Do not reuse a generic scene across story beats.
- Do not render a poster whose aspect ratio differs from the video canvas; `contain` would create side or top bars, while `cover` can remove important artwork.
- Do not let captions, narration, and scene meaning diverge.
- Do not use music without checking its current license and retaining attribution when required.
- Do not report a silent picture track as a completed narrated video.
- Do not expose intermediate files as final deliverables.
- Do not use a generated image's embedded text as poster copy; the renderer owns text layout.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The default prompt is written in English and instructs the model to 'Use $narrated-handdrawn-story-video...' despite the skill's display name and description being explicitly Chinese-focused. This creates a language/locale policy concern because it imposes a specific language context without offering user choice or documenting why English is required.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest sets "language": "Chinese" as a fixed requirement, which is a natural-language policy concern when no user opt-in or documented region-specific justification is present. This can restrict users to a single language without offering a choice.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README states the renderer '默认输出无配音、无音乐的 H.264 画面轨' and later repeats that output is silent and that voiceover/BGM are post-production work. This materially differs from the manifest, which claims the skill itself creates narrated videos with synchronized subtitles and licensed BGM mixing.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The line '输出是静音画面轨,配音和 BGM 属于后期工作' narrows the skill to silent visual rendering. The manifest instead advertises built-in Qwen3-TTS narration, optional character voices, subtitles, and music mixed under narration, which is a clear description-behavior mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The English overview says the project 'outputs a silent H.264 picture track ready for post-production voiceover,' which excludes built-in narration and music mixing. That directly conflicts with the manifest's description of generating polished narrated Chinese story videos with synchronized audio and subtitles.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The README explicitly says 'output is a silent picture track — voiceover and BGM are post-production' and the outputs table confirms silent MP4 files. This is not a minor implementation detail; it meaningfully conflicts with the manifest's claimed TTS narration, subtitles, and BGM mixing behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This code performs file-writing operations via ffmpeg with '-y' overwrite flags and copies a storyboard into place, which can replace existing outputs without user confirmation. While there is some post-action logging, there is no pre-action warning, confirmation prompt, or inline documentation disclosing that existing files will be overwritten when the script runs with these options.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This code creates an output directory and later copies files and writes manifests under project paths, which modifies the filesystem. While the behavior may be intended, the file itself provides no confirmation prompt and no comment/docstring warning the user that running the script will create files under public/assets and overwrite storyboard.uploaded.json and uploaded-pages.json.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest describes creating polished Chinese narrated short videos with Qwen3-TTS narration, subtitles, and licensed BGM mixing. This entrypoint only orchestrates image/story planning and renders outputs explicitly labeled as silent videos, with no code paths for narration, subtitle generation, or BGM mixing.

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 · renderer/scripts/run_story_video.py (reported line 79)May include surrounding context.

python
def run(command: list[str], project: Path) -> None:
    subprocess.run(command, cwd=project, check=True)


def main() -> None:

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The default title is hard-coded in Simplified Chinese, and later prompt construction and caption instructions require Simplified Chinese output. This imposes a specific language/locale on generated content without offering the user a language choice or documenting a justified regional constraint.

Content

No source excerpt is available for this finding.