ClawHub
Back to skill

Security audit

Clawie Research Agent

Security checks for vulnerabilities and agentic risk

Overview

This research skill is coherent and disclosed, but it sends research queries to public services, so users should avoid sensitive topics.

Install only if you are comfortable with research topics being sent to external search, GitHub, Hacker News, and npm services. Use it for public or non-sensitive research, and do not include secrets, proprietary identifiers, private customer names, or confidential internal project details in queries.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (6)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill advertises broad, structured, citation-backed research, but the implemented behavior is limited to a small set of shell-based lookups and raw result fetching. This mismatch is security-relevant because operators may overtrust the skill’s outputs and permissions profile, while hidden implementation limits and subprocess behavior can cause unsafe delegation, inaccurate conclusions, or misuse in contexts expecting controlled report generation.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill invokes shell-capable commands (curl, gh, npm, jq) but does not declare any tool scope or permissions boundary. This is dangerous because a caller or orchestrator may activate the skill without understanding it performs external network access and subprocess execution, increasing the risk of unintended data exposure or overly broad runtime privileges.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The activation criteria are extremely broad ('research, analysis, or investigation'), making the skill likely to trigger on many generic requests. In combination with shell and external network usage, this increases the chance of the skill being selected inappropriately, causing unnecessary external transmission of user-supplied topics or execution of tooling when a safer, narrower skill would suffice.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The skill sends query data to an external service (api.duckduckgo.com) via curl, which constitutes outbound data transmission. In a research skill this may be expected, but it is still dangerous if user prompts, proprietary topics, or internal identifiers are inserted into QUERY, because sensitive information could be disclosed to third parties without clear consent or sanitization.

Content

Scanner excerpt · SKILL.md (reported line 22)May include surrounding context.

bash
# Web search (DuckDuckGo)
curl -s "https://api.duckduckgo.com/?q=QUERY&format=json" | jq '.RelatedTopics[:5]'

# GitHub (repos, stars, activity)
gh search repos "TOPIC" --limit 20 --json name,description,stargazersCount,url

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/research.py (reported line 15)May include surrounding context.

python
def github_search(topic: str, limit: int = 20) -> list:
    """Search GitHub repos for a topic."""
    try:
        result = subprocess.run(
            ["gh", "search", "repos", topic, "--limit", str(limit),
             "--json", "name,description,stargazersCount,url,updatedAt,language"],
            capture_output=True, text=True, check=True

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/research.py (reported line 28)May include surrounding context.

python
def npm_search(package: str, limit: int = 5) -> list:
    """Search NPM for packages."""
    try:
        result = subprocess.run(
            ["npm", "search", package, "--json"],
            capture_output=True, text=True
        )

Static analysis

No suspicious patterns detected.