Back to skill

Security audit

图表工坊专业版

Security checks for vulnerabilities and agentic risk

Overview

The skill is a chart-generation helper, but its local execution, file-writing, dependency installation, network/API claims, and missing referenced scripts are not scoped clearly enough for automatic trust.

Review this skill before installing. Use it only in a dedicated workspace or virtual environment, avoid running dependency installs as administrator/root, confirm output paths before generation, and treat the network/API sections as unresolved until the publisher clarifies whether the tool is truly local-only and supplies the referenced implementation scripts.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:154
Finding

Unpinned Third-Party Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 154–156 and line 247
Vulnerability Type: Supply-chain risk from unpinned dependencies
Risk Level: Medium

Vulnerable Code

Lines 154–156:

bash
pip3 install matplotlib
pip3 install numpy
pip3 install pandas

Line 247:

bash
pip3 install matplotlib numpy pandas

Technical Analysis

The Skill directs users or execution-capable agents to install third-party Python packages without version constraints, cryptographic hashes, a lockfile, or an isolated environment. Package installation can execute package build and installation logic with the privileges of the invoking process.

Although the named packages are established projects and the audited file does not specify a malicious package or custom package index, resolving mutable package names at installation time creates a supply-chain exposure. A compromised release, compromised package registry, malicious index configuration, or unexpected future version could introduce unwanted code after the Skill has already been reviewed.

The document also claims that the tool operates entirely without network access, but these installation commands normally contact a package registry unless dependencies are already cached. This inconsistency may prevent users from recognizing that external artifacts are being retrieved.

Attack Path

  1. A user or agent follows the dependency-installation instructions.
  2. pip3 resolves the unpinned package names using the active Python package-index configuration.
  3. Package distributions and any required transitive dependencies are retrieved from mutable external sources or local mirrors.
  4. Installation or build hooks execute under the invoking account.
  5. If a resolved artifact, dependency release, configured mirror, or package-index channel has been compromised, attacker-controlled code executes locally.
  6. The resulting package remains install ...[truncated 909 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin every direct dependency to a reviewed version rather than installing unconstrained package names.
  2. Provide a lockfile that also fixes transitive dependency versions.
  3. Require cryptographic hashes, such as through a hash-locked requirements file and pip install --require-hashes.
  4. Install dependencies in a dedicated virtual environment instead of the global or system Python environment.
  5. Explicitly prohibit administrator or root installation and document a least-privilege setup procedure.
  6. Separate dependency provisioning from routine Skill execution so an agent cannot silently install packages during normal chart generation.
  7. Use a trusted, controlled package index or an internally reviewed artifact repository.
  8. Add dependency vulnerability and integrity scanning to the release process.
  9. Correct the documentation to disclose that initial dependency retrieval may require network access.
  10. Supply the referenced implementation scripts and a reproducible dependency manifest so reviewers can audit the actual runtime behavior.

Example hardened installation workflow:

bash
python3 -m venv .venv
.venv/bin/python -m pip install --require-hashes -r requirements.lock
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The description says to use the skill whenever users need data analysis, report generation, statistical insight, or data visualization, which is a very wide set of everyday requests rather than a narrow trigger. It does not provide explicit trigger phrases, activation boundaries, or negative examples beyond one excluded case, so the invocation scope is ambiguous.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This line states that the skill's core capability includes Chinese interaction, and the display name and surrounding documentation are also Chinese-centric. There is no indication that users may choose another language or that the Chinese-only behavior is a necessary region-specific constraint, which can violate language/locale policy expectations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill describes automated generation and archival of files into user-specified directories without clear warnings about write locations, overwrite behavior, or path safety. In an agent context with write and exec capabilities, this can lead to unintended modification of local files, clobbering important data, or writing artifacts to sensitive paths if user-controlled paths are passed through unchecked.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill exposes local shell execution examples and advertises exec/write tooling without a clear safety notice that commands run on the host and can create or overwrite files. In practice, users may treat examples as harmless data operations while the agent executes local commands, increasing the risk of accidental filesystem changes or unsafe parameterization of command arguments.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The runtime and API key sections explicitly state that the skill runs completely locally, needs no network access, and uses no external services. However, later troubleshooting and security sections reference API keys, 401 authentication failures, 429 rate limits, 504 responses, DNS failures, HTTPS, and certificate validation, which directly contradict the earlier stated operating model.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The manifest description states the skill is not applicable to real-time stream data processing. Later, the limitations section says '不支持流式输出(需要专业版)', which suggests streaming is an available or planned capability gated by edition rather than outside the skill's scope, creating an intent inconsistency in the documentation.

Content

No source excerpt is available for this finding.