Back to skill

Security audit

避雷.skill

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign with medium confidence: the scans report no AIG findings, but flag an unpinned installer and an update checker outside the stated shopping-advisor purpose.

The verdict is benign, but it does not mean risk-free. Before installing, consider the unpinned installer, the update checker outside the stated purpose, and the reported cost-calculation issue. The scan is incomplete: 9 of 10 components were scanned, and static tool-misuse analysis was degraded, so the evidence does not establish that all behavior is safe.

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
  • 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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (10)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The supplied chunk's entire function is software-version maintenance, a separate capability not disclosed in the shopping-advisor description. Its network requests and cache support that update-checking function, not the declared product-research workflow. This assessment is limited to the supplied chunk and does not establish whether other files implement the advertised shopping features. The code does not automatically install updates, execute downloaded content, or visibly upload local file contents.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The installation command runs an npm-distributed CLI without pinning its version, exposing installations to changes or compromise of the package resolved at execution time. This is a supply-chain risk rather than an MCP-server issue: a malicious installer release could execute with the installing user's privileges.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The documented npx command executes an unpinned npm CLI, so the code used for installation can change without a corresponding documentation change. Although it is not an MCP-server reference, compromise of the resolved installer release could lead to arbitrary code execution under the user's account.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_write, network.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · scripts/check_update.py (reported line 22)May include surrounding context.

python
import urllib.request

SOURCES = [
    ("github-raw", "https://raw.githubusercontent.com/zhashushu/bilei-skill/main/SKILL.md"),
    ("jsdelivr", "https://cdn.jsdelivr.net/gh/zhashushu/bilei-skill@main/SKILL.md"),
]
CACHE_DIR = os.path.expanduser("~/.cache/bilei-skill")

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Warranty-period repair charges contradict the documented post-warranty cost model.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The exposure calculation allocates the total number of uninsured years to the beginning of ownership rather than to the period after warranty expiration. With a positive discount rate this overstates repair present value and can distort cost rankings and purchasing recommendations; at zero discount the aggregate repair cost can remain correct, masking the defect in existing tests.

Content

No source excerpt is available for this finding.

The trigger list includes generic terms such as “对比” (compare), “扫雷” (minesweeper/risk screening), and “梯队” (tiers) without requiring a shopping-related context.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

These terms commonly occur in requests unrelated to purchasing, such as comparing documents, playing Minesweeper, or ranking sports teams. Although the skill is described as a shopping advisor, the listed triggers do not explicitly require product-selection or purchase intent, creating ambiguity about when it should activate.

Content

No source excerpt is available for this finding.

The skill implements an undeclared self-update check unrelated to product vetting.

Low
Category
Excessive Agency
Confidence
92% confidence
Finding

The manifest describes shopping research, evidence grading, manufacturer audits, and quantitative comparisons, not skill-maintenance networking. This function contacts the skill publisher's GitHub/CDN sources to retrieve a remote skill version, and main invokes it when no usable cache exists (L120–L130); these requests do not gather shopping evidence. The code only checks versions and does not install or execute remote content.

Content

No source excerpt is available for this finding.