Back to skill

Security audit

AI科研工具全景指南

Security checks for vulnerabilities and agentic risk

Overview

This research-tools guide is mostly coherent, but it tells users to run unpinned remote installer commands that can change over time and install additional skills with unspecified provenance.

Review this skill before installing. It is a guide rather than a bundled program, but do not run the listed `npx ... @latest` commands unless you trust the package source, have checked the resolved versions and permissions, and are comfortable installing those extra skills into your agent environment. Avoid using confidential manuscripts, restricted datasets, reviewer materials, or personal data with external research connectors unless you have approval and have minimized or anonymized the data.

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:231
Finding

Unpinned Third-Party Package and Skill Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:231-254
Vulnerability Type: Supply-chain risk from mutable, unverified dependencies
Risk Level: Medium

The quick-start guide instructs users to execute the latest available version of a registry package and use it to install four additional Skills:

bash
# Install MCP tools
npx clawhub@latest install mcp-tools

# Configure academic data sources
openclaw mcp add academic --connector arxiv --connector semantic-scholar

# Install Stata Skill
npx clawhub@latest install stata-skill

# Translate using Python
stata-to-python --input paper.do --output paper.py

# Install the review-response assistant
npx clawhub@latest install review-response

# Convert a paper to PPT
npx clawhub@latest install paper-slide-deck

Technical Analysis

The four installation commands invoke npx clawhub@latest. The @latest tag is mutable and does not identify the exact package version or integrity hash reviewed by the author. npx may download and execute registry-hosted package code as the current user. The invoked installer then retrieves additional Skills whose versions, publishers, contents, integrity values, and required permissions are not specified in this artifact.

Consequently, the effective code and instructions executed by users can change after this Skill has been reviewed. A compromised registry account, package takeover, malicious future release, or compromised transitive dependency could introduce hostile behavior without requiring any modification to SKILL.md.

The MCP configuration and stata-to-python commands are included in the complete workflow snippet for context. The confirmed issue is specifically the use of mutable @latest installation commands and the absence of dependency verification.

Attack Path

  1. An attacker compromises the clawhub package, one of the named Skills, or a relevant dependency or publisher account.
  2. The at ...[truncated 1294 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace @latest with an exact, reviewed package version.
  2. Pin each installed Skill to a specific immutable version or commit identifier where the tooling supports it.
  3. Verify package integrity using registry checksums, lockfiles, signatures, provenance attestations, or trusted release hashes.
  4. Document the expected publisher, canonical repository, reviewed version, required permissions, and relevant integrity value for every dependency.
  5. Review the package, its lifecycle scripts, transitive dependencies, and each downloaded Skill before execution.
  6. Disable or avoid unnecessary installation scripts and execute installation in a sandbox or least-privilege environment.
  7. Add a confirmation step that displays the resolved version and source before downloading or executing it.
  8. Maintain an allowlist of approved packages and Skills, monitor them for ownership or release changes, and document rollback and removal procedures.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The workflow describes automatic data retrieval, analysis, and report generation but does not warn about privacy, data governance, or handling of uploaded research materials. In a research setting, this can expose unpublished data, personally identifiable information, licensed datasets, or sensitive policy materials to external systems without informed consent or compliance checks.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The skill instructs users to run npx clawhub@latest install mcp-tools, which fetches and executes the latest package code without version pinning. In a skill context, this is a supply-chain risk: a compromised upstream package, malicious update, or typosquatted dependency could lead to arbitrary code execution on the user's machine.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The command npx clawhub@latest install stata-skill executes unpinned remote code. Because this skill is aimed at researchers who may run commands directly, the lack of version pinning materially increases exposure to supply-chain compromise and unintended code changes over time.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The skill recommends npx clawhub@latest install review-response, again invoking unpinned package execution. This creates a repeatable supply-chain weakness where future package updates can silently alter behavior or execute malicious payloads under the user's privileges.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The command npx clawhub@latest install paper-slide-deck has the same unpinned execution risk as the other install examples. In aggregate, repeated unsafe install guidance normalizes executing mutable third-party code and broadens the attack surface for users following the skill verbatim.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger examples are broad, covering common phrases like helping with literature reviews, reproducing papers, replying to reviewers, or asking whether AI can help with research. Without scope constraints, the skill may activate in contexts containing sensitive unpublished manuscripts, confidential datasets, or regulated research material, causing over-collection, unwanted tool use, or inappropriate delegation to external services.

Content

No source excerpt is available for this finding.