Back to skill

Security audit

B 站数据助手

Security checks for vulnerabilities and agentic risk

Overview

ClawScan verdict: benign, with medium confidence; the scanners report a public Bilibili data analysis skill with some undisclosed local recordkeeping and other limited concerns, while the scan is incomplete.

The verdict is benign, not a guarantee of safety, and confidence is medium. Before installing, consider whether the local confirmation and audit records are acceptable, whether account-balance access is needed, and whether you can use a reviewed, pinned installer version. Skillspector marked its analysis partial because some local references were missing or ambiguous, so the evidence is incomplete.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (8)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Local bookkeeping can be a supporting implementation detail, but this code goes beyond neutral logging: it unconditionally asserts that legal confirmation occurred. No confirmation interaction or validation appears in the supplied chunk. That consent-recording behavior is not disclosed by the public-data query and analysis description. The mismatch is not merely that this excerpt lacks API calls.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skillhub'.

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

The installation command runs an unpinned npm package through npx, allowing the executable installer to change between installations. A compromised future release could execute with the user's local permissions and access files or credentials; the finding's MCP-server characterization is inaccurate, but the underlying supply-chain exposure is real.

Content

No source excerpt is available for this finding.

The whitelist permits querying the MaxHub account balance, outside the stated Bilibili public-data analysis scope.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The maxhub_get_balance entry authorizes GET requests to /api/maxhub/balance, an account-balance operation rather than a Bilibili public-data query. Although it uses the declared host and is read-only, accessing service-account financial information is not justified by the manifest's stated content-analysis capabilities.

Content

No source excerpt is available for this finding.

The mandatory consent gate only accepts Chinese-language confirmations, without offering a language choice.

Low
Category
Output Handling
Confidence
96% confidence
Finding

The acceptance rules whitelist only Chinese consent phrases and explicitly reject English responses such as “yes.” Because users cannot proceed without passing this gate, the skill imposes a Chinese-language interaction requirement without user opt-in or a documented justification, despite declaring English locale support.

Content

No source excerpt is available for this finding.

The no-write constraint contradicts mandatory local state and audit-log writes.

Low
Category
Excessive Agency
Confidence
89% confidence
Finding

L250 explicitly states that the skill performs only queries and analysis and does not execute writes. However, L315–L321 require creating a directory, overwriting a persistent confirmation file, and appending an audit log under $HOME; L422–L426 additionally require logging every successful API call. These are bounded compliance-related writes, not remote Bilibili modifications, but the unconditional no-write statement does not disclose that distinction.

Content

No source excerpt is available for this finding.

Generic triggers can activate a Bilibili-specific recipe for unrelated link-sharing or profile requests.

Low
Category
Trigger Abuse
Confidence
88% confidence
Finding

The activation phrases include “分享链接” (share a link) and “看主页” (view a profile), which apply to many platforms and everyday requests. Although the input requires a https://b23.tv/ link, the trigger description does not require that context before selecting this recipe.

Content

No source excerpt is available for this finding.

Batch-operation reminders use broad everyday keywords without requiring a data-collection context.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

Keywords such as “全部” (all), “所有” (every), and “循环” (loop) can occur in ordinary requests that do not involve bulk collection, such as asking to explain all endpoint parameters. Although the category description mentions bulk data collection, the matching configuration does not require that context or provide exclusions.

Content

No source excerpt is available for this finding.

Confirmation rules require Chinese wording without offering a language choice.

Low
Category
Output Handling
Confidence
96% confidence
Finding

Every accepted confirmation is Chinese, while “yes” is explicitly rejected and the fallback prompt directs users to enter Chinese wording. This imposes a language-specific interaction requirement without user opt-in or a documented justification for restricting confirmation to Chinese.

Content

No source excerpt is available for this finding.