ClawHub
Back to skill

Security audit

autonomous-deep-research

Security checks for vulnerabilities and agentic risk

Overview

This skill is a local research helper with disclosed report generation and optional local learning state, with no evidence of exfiltration, credential use, hidden execution, or destructive behavior.

Before installing, be aware that reports and optional learning data are written locally, and the learning helper can retain preferences, errors, and notes in learned_patterns.json. Avoid storing sensitive information in those notes or preferences unless you are comfortable with that local persistence.

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (4)

Lp3

Medium
Category
MCP Least Privilege
Confidence
87% confidence
Finding
The skill documentation advertises shell execution and file-writing behavior without declaring corresponding permissions, which breaks least-privilege expectations and makes it harder for a host to constrain risky operations. In an autonomous research skill, undeclared write/command capabilities could be abused to modify local files, persist data, or invoke external programs outside the user’s intent.

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding
The declared purpose is research/synthesis, but the documentation also describes persistence of usage history, user preferences, error patterns, and automated self-optimization behavior. This mismatch is dangerous because operators may grant or invoke the skill for research while not realizing it also performs ongoing profiling and stateful data retention, increasing privacy and integrity risk.

Description-Behavior Mismatch

Medium
Confidence
91% confidence
Finding
The file explicitly presents itself as a generic cross-skill 'self-evolution' module that any WorkBuddy skill can call, which exceeds the declared purpose of a deep-research skill. This creates an unnecessary persistence and telemetry layer across skills, increasing privacy and governance risk because unrelated skills can reuse it to accumulate behavior history and operational metadata without clear scoping or consent boundaries.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The code records capability usage, failures, free-form notes, recurring errors, and arbitrary user preferences into a persistent JSON file, providing reusable profiling and behavior-tracking functionality not clearly required for answering research questions. Because notes and preferences are unconstrained, the module can retain sensitive user or workflow information and make that data available for later reuse by the skill, creating privacy, data-minimization, and covert statefulness concerns.

Static analysis

No suspicious patterns detected.