Back to skill

Security audit

Song Creation

Security checks for vulnerabilities and agentic risk

Overview

The skill appears intended for song generation, but it needs review because it installs mutable third-party components and contains unsafe local file-moving behavior.

Install only after reviewing the setup script and preferably running it in an isolated account, VM, or container. Pin and verify ComfyUI, custom nodes, Python dependencies, and model downloads before use; remove broad CORS unless needed; avoid untrusted song names; and fix path validation before allowing automated output moves.

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 (5)

T08 · Insecure Dependencies

Error
Location
setup.sh:77
Finding

Unpinned executable dependencies permit supply-chain compromise

Content
View full analysis
/dev/null || true fi done ``` ### Technical Analysis The setup process clones the current default branches of ComfyUI and ComfyUI-Manager without pinning immutable commit hashes. It then installs the cloned repository's requirements and directly installs several Python packages without a hash-locked dependency manifest. Consequently, the code executed by a future installation is not fully represented by the audited project. A compromised upstream repository, malicious package release, dependency takeover, or unexpected upstream change could introduce arbitrary Python code. Suppressing ComfyUI-Manager clone errors with `|| true` also makes installation state harder to verify reliably. ### Attack Path 1. An at ...[truncated 903 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
setup.sh:125
Finding

Downloaded model artifacts are not pinned or integrity-verified

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/submit_song.py:102
Finding

Unsanitized song names allow output path traversal

Content
View full analysis
/audio`. 5. `shutil.move()` places the generated MP3 at the unintended user-writable location and may overwrite an existi ...[truncated 453 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/submit_song.py:140
Finding

Untrusted ComfyUI output metadata can move arbitrary accessible files

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
setup.sh:168
Finding

Permissive CORS exposes the local ComfyUI API to arbitrary web origins

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (19)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

If the skill really performs environment setup, installs software, downloads external models, clones remote repositories, and creates local scripts beyond the declared song-creation workflow, then the mismatch conceals materially riskier behavior from the user and reviewer. Hidden installation and code-fetching behavior is dangerous because it expands execution and supply-chain exposure under a benign-seeming description.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

If the skill really performs environment setup, installs software, downloads external models, clones remote repositories, and creates local scripts beyond the declared song-creation workflow, then the mismatch conceals materially riskier behavior from the user and reviewer. Hidden installation and code-fetching behavior is dangerous because it expands execution and supply-chain exposure under a benign-seeming description.

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/submit_song.py (reported line 119)May include surrounding context.

python
print(f"Submitted: {prompt_id}")

    if not wait:
        return prompt_id

    # Poll until completion
    for i in range(600):

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill declares broad operational behavior that includes reading local files, invoking shell commands, and using network-backed models, but it does not declare any explicit tool scope or least-privilege restrictions. In practice, this increases the blast radius of prompt injection or misuse because a song-generation request can trigger filesystem access, subprocess execution, and remote API usage without clear containment.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger phrase "any song/music creation request" is overly broad for a skill that can read files, spawn subprocesses, move/delete files, and interact with external tools. Broad invocation criteria make accidental activation more likely, causing privileged actions to run in contexts where the user may have expected only conversational help.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill performs potentially impactful side effects—starting ComfyUI, creating files, moving outputs, deleting directories, and managing processes—without clear user-facing disclosure. Lack of transparency undermines informed consent and increases the risk of unexpected local changes from a request that sounds like simple content generation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The task template hard-codes the synthesis language as zh, which imposes a specific locale without indicating that the user can choose another language. This is a natural-language locale policy issue because the skill does not document opt-in or present language selection.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction to run pkill -f "python.*main.py" gives the skill host-level process-management power that can terminate unrelated user workloads, not just ComfyUI. In the context of a media-generation skill, this is overbroad and can cause denial of service or accidental interruption of other local applications.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The function sets language: str = "zh" as the default, which imposes a specific language/locale behavior unless the caller overrides it. The policy allows locale constraints only when users are given a choice or the constraint is clearly justified; this file does not present such a user opt-in or justification.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The script transmits user-provided lyrics and metadata to an external service endpoint via HTTP. Although the target is localhost, this still crosses a process boundary and may expose sensitive user content to another service, logs, plugins, or a misconfigured network binding; the skill context increases relevance because lyrics may contain private or copyrighted material.

Content

Scanner excerpt · scripts/submit_song.py (reported line 108)May include surrounding context.

python
wf["107"]["inputs"]["filename_prefix"] = prefix

    # Submit
    resp = requests.post(f"{COMFYUI_API}/api/prompt", json={"prompt": wf})
    data = resp.json()
    prompt_id = data.get("prompt_id")
    errors = data.get("node_errors", {})

Tainted flow: 'prompt_id' from requests.post (line 110, network input) → requests.get (network output)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/submit_song.py (reported line 124)May include surrounding context.

python
# Poll until completion
    for i in range(600):
        try:
            resp = requests.get(f"{COMFYUI_API}/api/history/{prompt_id}")
            h = resp.json().get(prompt_id, {})
            if h.get("status", {}).get("completed"):
                break

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The command-line interface defines --lang with a default of zh, which means the skill will produce Chinese-language behavior even when the user did not request a specific locale. This is a natural-language policy concern because it fixes language output without explicit opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This shell script presents its user-facing workflow in Chinese, and later instructs users to trigger the skill with the fixed Chinese phrase "创作一首歌." That constitutes a language/locale constraint in natural-language content without offering any language choice or documenting a justified region-specific limitation.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · setup.sh (reported line 181)May include surrounding context.

sh
# ─── CREATE: engine operation manual ─────────────────────
header "生成引擎操作手册"
MANUAL="$SKILL_DIR/references/comfyui_music_generation.md"
mkdir -p "$(dirname "$MANUAL")"
cat > "$MANUAL" << 'MANUAL'
# ComfyUI + AceStep 音乐生成操作手册(引擎 ⚙️ 专用)

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · setup.sh (reported line 181)May include surrounding context.

sh
# ─── CREATE: engine operation manual ─────────────────────
header "生成引擎操作手册"
MANUAL="$SKILL_DIR/references/comfyui_music_generation.md"
mkdir -p "$(dirname "$MANUAL")"
cat > "$MANUAL" << 'MANUAL'
# ComfyUI + AceStep 音乐生成操作手册(引擎 ⚙️ 专用)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The generated markdown manual says it is "专用" and all operational instructions are written only in Chinese. This is a natural-language policy issue because it enforces a specific language for operators without any opt-in, fallback, or explanation that the skill is region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The usage summary instructs users to invoke the skill by saying the Chinese phrase "创作一首歌" and provides no alternative trigger wording in other languages. This creates an undocumented locale-specific activation requirement, which violates the language/locale policy criteria.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
75% confidence
Finding

The manifest says the skill creates songs through a writer → musician → programmer workflow, but the documented implementation requires reading ~/.openclaw/workspace/tools/comfyui操作手册.md from disk. Accessing arbitrary local documentation is not part of the user-facing purpose and introduces a broader file-read capability than the manifest explains.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
98% confidence
Finding

The submit() docstring documents the return value as "MP3 文件路径,或 None", implying callers receive a file path after generation. However, the implementation returns prompt_id directly when wait is False, which is a different kind of value and changes the function's observable behavior.

Content

No source excerpt is available for this finding.