Back to skill

Security audit

皮皮虾数据助手

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report undeclared local compliance records, update guidance, and an unpinned installer; the evidence is incomplete because reference resolution was partial.

Consider the reported undeclared persistent records, upgrade instructions, and unpinned installer before installing. The findings support a suspicious review verdict, not a claim of malicious intent. Scanning covered the listed components, but reference resolution was partial, 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
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (18)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose concerns content-data retrieval, but this chunk records a confirmed legal status and a first-time-confirmation event. These writes assert user confirmation rather than merely supporting API access through ordinary configuration or diagnostic logging. That compliance-recording behavior is not disclosed in the description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Ordinary operational logging could support the declared query functionality, but this code specifically records a completed legal review without performing or verifying one in the supplied chunk. That is a distinct compliance-related action rather than logging a content query. No network transmission or data exfiltration is shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

Routine logging could be a supporting implementation detail, but this code also unconditionally marks legal version 1.0.0 as confirmed and records a first-time confirmation event. Establishing that consent-related state is a substantive behavior not disclosed by the content-query description. The mismatch is this affirmative confirmation recording, not merely the absence of API calls in a partial code chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The skill recommends package upgrades, including a forced upgrade, in response to API errors without a separate explicit approval or version-verification requirement. This can turn a read-only query into installation of changed agent instructions from another service; however, the finding's claimed skillhub install --force command is not present in this source.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

Ordinary operational logging could be a supporting implementation detail. This chunk instead records a specific legal-review completion without showing any review or content-query operation. That compliance-state assertion is distinct from the declared querying capability. The append succeeds only if the destination directory exists and is writable.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The installation command executes an unpinned npm package through npx, so the executable obtained can change without any change to this documentation. A compromised publisher or malicious future release could execute code with the installing user's permissions; this is an installer supply-chain risk, not specifically an MCP-server issue.

Content

No source excerpt is available for this finding.

The mandatory consent gate requires Chinese confirmation phrases without offering a language choice.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The acceptance rules list only Chinese confirmation phrases and explicitly reject English responses such as 'yes' and 'ok'. Because the skill cannot proceed without passing this gate, it imposes a Chinese-language interaction requirement without user opt-in or a documented locale-specific justification.

Content

No source excerpt is available for this finding.

The read-only query skill requires persistent writes to the user's home directory.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest describes content-data queries and declares capability: read_only (L011), but this executable workflow creates a directory, overwrites a confirmation file, and appends an audit log under $HOME/.maxhub-skills/.compliance. Further mandatory logging records each successful API call (L417–L421), making persistent local modification part of normal operation rather than an optional user-requested export.

Content

No source excerpt is available for this finding.

API failures can trigger skill-package upgrades outside the declared query-only behavior and network destination.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

Error-handling guidance recommends modifying the installed skill through unpinned upgrade commands, including --force, without requiring a separate maintenance confirmation. If an agent interprets these recommendations as recovery actions, an ordinary query failure can cross into package replacement and a different network trust boundary.

Content

No source excerpt is available for this finding.

The search workflow sends user-provided keywords to a third-party API without a privacy warning.

Medium
Category
Output Handling
Confidence
85% confidence
Finding

The usage instructions direct searches using user-provided keywords, and the workflow at L032 explicitly chains fetch_search into a detail lookup. Requests go to www.aconfig.cn (L003), but the file does not warn users that their search terms will be disclosed to that third-party service rather than sent only to PiPiXia.

Content

No source excerpt is available for this finding.

Recipe triggers include everyday phrases without a platform or task scope restriction.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

L15 lists "数据" (data), and L16 lists "有什么好看的" (anything good to watch/see) as activation keywords. These can match unrelated analytics or recommendation requests; the longest-match and tie-breaking rules at L5–L6 do not require confirming that the user intends a 皮皮虾 recipe.

Content

No source excerpt is available for this finding.

The lookup workflows send user identifiers or search terms to a third-party API without a privacy disclosure.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The chain recipes direct user-profile and social-graph lookups through the API hosted at aconfig.cn (L003), transmitting user IDs or search terms to a service separate from PiPiXia. These are actionable workflows rather than endpoint listings, but the file does not warn users that their lookup inputs will be shared with that third party.

Content

No source excerpt is available for this finding.

L144–L149 authorize querying the MaxHub account balance, a capability outside the stated Pipixia content-query purpose.

Low
Category
Excessive Agency
Confidence
86% confidence
Finding

The whitelist permits GET /api/maxhub/balance, which retrieves provider-account information rather than Pipixia content or engagement data. Account-balance inspection is not declared in the manifest and is not a direct requirement of the described research and content-query functions.

Content

No source excerpt is available for this finding.

The statistics recipe includes the overly broad trigger “数据” (“data”).

Low
Category
Trigger Abuse
Confidence
98% confidence
Finding

“数据” occurs in many requests unrelated to post statistics. The activation rule does not explicitly require a Pipixia post context, so unrelated data requests could select this recipe.

Content

No source excerpt is available for this finding.

The home-feed recipe uses generic recommendation requests as activation triggers.

Low
Category
Trigger Abuse
Confidence
98% confidence
Finding

“推荐” (“recommend”) and “有什么好看的” (“anything good to watch?”) can refer to many services and content types. Listing them alongside “皮皮虾热门” does not explicitly restrict those generic alternatives to Pipixia.

Content

No source excerpt is available for this finding.

The trigger “社交圈” (“social circle”) is broad enough to activate the recipe during ordinary discussion of social relationships.

Low
Category
Trigger Abuse
Confidence
88% confidence
Finding

The recipe lists “社交圈” as a semantic activation condition without requiring a request to retrieve a platform user's followers or following list. This could invoke the data-retrieval workflow for unrelated conversation about someone's social circle.

Content

No source excerpt is available for this finding.

Batch-operation triggers include everyday words without requiring a bulk-data-collection context.

Low
Category
Trigger Abuse
Confidence
94% confidence
Finding

The keyword list includes 全部 ('all'), 所有 ('every/all'), and 循环 ('loop'), which can occur in ordinary requests unrelated to bulk data collection. Although the category description names bulk collection, the matching rules do not explicitly require that context or provide exclusions, allowing unrelated requests to trigger a compliance interruption.

Content

No source excerpt is available for this finding.

Confirmation rules require Chinese-language responses 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 the user to enter 同意. This restricts the user's confirmation language without opt-in or a documented justification; the issue is the enforced response vocabulary, not the language of the documentation.

Content

No source excerpt is available for this finding.