Back to skill

Security audit

Deepsop Bilibili 工作台

Security checks for vulnerabilities and agentic risk

Overview

This skill does what it says, but it asks the agent to automatically download and run unverified third-party upload software and fallback mirrors with too little user control.

Review this skill carefully before installing. Only use it if you are comfortable with OpenClaw cloning and running the external social-auto-upload project from network sources, storing Bilibili login state locally, and uploading selected files and metadata to Bilibili. Prefer a version that pins reviewed commits/releases, verifies checksums or signatures, removes automatic mirror fallback, and asks before setup or upload actions.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (2)

T03 · Remote Payload Retrieval and Execution

Error
Location
references/runtime-requirements.md:35
Finding

Automatic Execution of Mutable Code Retrieved Through Untrusted Mirrors

Content
View full analysis
git clone https://gh-proxy.org/https://github.com/dreammis/social-auto-upload.git git clone https://gh-proxy.com/https://github.com/dreammis/social-auto-upload.git git clone https://hub.gitmirror.com/https://github.com/dreammis/social-auto-upload.git ``` From `references/runtime-requirements.md:81-100`: ```bash cd copy conf.example.py conf.py cp conf.example.py conf.py uv sync --python 3.12 uv run --project python sau_cli.py --help ``` ### Technical Analysis The Skill requires the agent to clone the `social-auto-upload` repository from a mutable default branch and then install its dependencies and execute its Python entry point. No immutable Git commit, signed tag, trusted checksum, or repository signature is verified before execution. If direct GitHub access fails, the instructions require automatic retries through several third-party services. One fallback is a repository under a different owner: ```text Declared upstream: github.com/dreammis/social-auto-upload Fallback source: gitee.com/lonycn/social-auto-upload ``` This is not merely an alternate transport endpoint. It is a separately controlled repository whose equivalence to the declared upstream is not established. The proxy services also terminate or relay retrieval through additional in ...[truncated 2570 chars]
Remediation
View remediation

T03 · Remote Payload Retrieval and Execution

Error
Location
references/troubleshooting.md:18
Finding

Executable Archive Downloaded Through Third-Party Proxy Without Integrity Verification

Content
View full analysis
/ https://gh-proxy.com/https://github.com/biliup/biliup/releases/download// ``` The same document states at `references/runtime-requirements.md:142` that the uploader automatically downloads the `biliup` binary during the first video upload. ### Technical Analysis The Skill directs users toward third-party proxy domains when downloading a native executable archive. It does not provide an expected cryptographic digest, signature, trusted signing key, or verification procedure. TLS protects only the connection to the proxy domain. It does not prove that the returned archive is identical to the artifact published by the `biliup` maintainers. If a proxy is compromised or malicious, it can substitute a modified archive while still presenting a valid certificate for its own domain. The troubleshooting URL pins a version number, but version pinning alone does not ensure integrity. An attacker controlling the delivery path could return arbitrary bytes for that versioned URL. The broader runtime instructions also use `` and `` placeholders and do not require a reviewed version. Because `biliup` is an executable component used by the upload workflow, substitution can result in native code execution. This retrieval is related to the declared upload functionality, but accepting it without cryptographic verification exceeds the minimum trust necessary. ### Attack Path ...[truncated 1581 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
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (14)

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · references/cli-contract.md (reported line 1)May include surrounding context.

md
# 假设 social-auto-upload 已 clone 到 $SAU_HOME,且已经跑过 `uv sync --python 3.12`。
$SAU_HOME = "$env:USERPROFILE\.openclaw\social-auto-upload"

# 登录

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · references/runtime-requirements.md (reported line 1)May include surrounding context.

md
# 假设 social-auto-upload 已 clone 到 $SAU_HOME,且已经跑过 `uv sync --python 3.12`。
$SAU_HOME = "$env:USERPROFILE\.openclaw\social-auto-upload"

# 登录

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · references/troubleshooting.md (reported line 1)May include surrounding context.

md
# 假设 social-auto-upload 已 clone 到 $SAU_HOME,且已经跑过 `uv sync --python 3.12`。
$SAU_HOME = "$env:USERPROFILE\.openclaw\social-auto-upload"

# 登录

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · scripts/examples/bilibili_commands.ps1 (reported line 1)May include surrounding context.

text
# 假设 social-auto-upload 已 clone 到 $SAU_HOME,且已经跑过 `uv sync --python 3.12`。
$SAU_HOME = "$env:USERPROFILE\.openclaw\social-auto-upload"

# 登录

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The README describes uploading videos to Bilibili via an external CLI but does not provide a clear privacy or network warning that user-supplied media and account actions will be transmitted to a third-party service through external tooling. In this skill context, that omission is material because the core function is handling potentially sensitive video content and authentication state, so users need explicit notice before data leaves the machine.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README explicitly states that the tool will automatically clone a repository and prepare dependencies on first use, but it does not clearly warn users that this will modify the local system and execute code obtained from an external project. That creates a supply-chain and local-environment risk because users may invoke the skill without understanding that it performs installation-like actions beyond the immediate upload task.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill instructs the agent to perform shell-capable actions such as cloning a repository, syncing dependencies, and running Python commands, but it does not declare any explicit tool scope or allowed-tools/permissions metadata. This creates an authorization gap where the agent may execute impactful local and networked operations without clear least-privilege boundaries or user-visible consent controls.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The upload contract instructs the agent to send a local file, title, description, category, and optional tags/schedule to Bilibili, but it does not explicitly warn that this causes external transmission of local content and metadata. In an agent setting, missing this disclosure can lead to unintended exfiltration of user-controlled files or sensitive metadata because the action is operationally significant and outward-facing.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file tells the agent to automatically prepare the environment, including cloning a repository and synchronizing dependencies, without first obtaining clear user consent for filesystem and network changes. Silent installation behavior is dangerous because it normalizes unannounced modification of the user's environment and can mask execution of untrusted code.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions explicitly tell the agent to perform silent existence checks, cloning, and dependency setup before user-facing interaction, and even forbid asking the user during retry behavior. This encourages hidden network and filesystem actions, increasing the risk of unauthorized system modification and making it harder for users to assess what external code is being fetched and executed.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill instructs the agent to clone and execute a broad third-party project, social-auto-upload, which supports multiple platforms and performs code execution outside the narrow Bilibili-only scope advertised by the skill. Pulling and running external source code materially expands the trust boundary and creates supply-chain and overreach risk if the repository or its dependencies are compromised.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The documentation mandates automatic fallback to multiple third-party mirror and proxy domains for source retrieval, which weakens provenance guarantees and increases the chance of downloading tampered code. Because these mirrors are outside the primary publisher-controlled distribution path, they introduce significant supply-chain risk for arbitrary code execution on the user's machine.

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 · scripts/examples/bilibili_cli_template.py (reported line 36)May include surrounding context.

python
"--tags", "足球,测试",
        "--schedule", "2026-03-26 16:00",
    ]
    subprocess.run(command, check=True)


if __name__ == "__main__":

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The entire operational guidance is presented only in Chinese, and there is no indication that users may choose another language or that the locale restriction is required for a region-specific compliance reason. This can violate a language/locale policy when the skill effectively forces one language without opt-in.

Content

No source excerpt is available for this finding.