ClawHub
Back to skill

Security audit

Pilot Dns

Security checks for vulnerabilities and agentic risk

Overview

This skill is a small, disclosed Pilot Protocol naming helper, with one low-risk example that goes beyond naming by sending a fixed test message to another agent.

Install only if you intend to let the agent use `pilotctl` for Pilot Protocol naming tasks. Review any proposed `connect` command before allowing it, since that can contact another agent even though the main purpose of the skill is hostname management.

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
Findings (1)

other

Note
Location
SKILL.md:74
Finding

Out-of-Scope Outbound Agent Communication

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 74-77
Vulnerability Type: other: Unnecessary outbound agent communication
Risk Level: Low

bash
# Resolve specific worker
worker_id=$(pilotctl --json find "ai-worker-02" | jq -r '.node_id')

# Connect using hostname
pilotctl --json connect "ai-worker-02" 7 --message "Hello"

Technical Analysis

The skill is documented as a human-friendly naming system for setting hostnames, resolving names and node IDs, and listing peers. However, its workflow example goes beyond those responsibilities by instructing the agent to connect to a remote peer and transmit a message.

Establishing an outbound connection is not required to demonstrate naming or lookup functionality. An AI agent following the example as an operational workflow could therefore initiate network communication without the user explicitly requesting it. The preceding lookup assigns a node ID to worker_id, but that variable is not used by the connection command.

The reviewed content transmits only the fixed string Hello; it does not collect or send credentials, files, environment variables, or other sensitive data. The external pilotctl implementation was not included in the project and could not be audited, so no broader behavior is asserted.

Attack Path

  1. The agent loads the pilot-dns skill to perform a naming-related task.
  2. The agent follows the documented workflow example.
  3. pilotctl --json find "ai-worker-02" resolves the target hostname.
  4. pilotctl --json connect "ai-worker-02" 7 --message "Hello" initiates communication with that peer.
  5. The fixed message is transmitted even though outbound messaging is outside the skill's stated naming purpose.

Impact Assessment

Successful execution permits an outbound connection and transmission of the supplied message under the permissions and network access already available to pilotctl. In the demonstrated ca ...[truncated 398 chars]

Remediation
View remediation

Remediation Suggestions

  • Remove the pilotctl connect command from this naming-focused skill.
  • End the workflow after hostname resolution, or print the resolved node identifier without initiating communication.
  • Move connectivity and messaging examples into a dedicated networking or communication skill.
  • Require explicit user confirmation before contacting a remote peer or transmitting any message.
  • If a cross-skill example is retained, clearly label it as optional and state its network side effects.
  • Remove the unused worker_id assignment or use it only in a non-networking demonstration.

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is declared as a naming and alias-management capability, but the example workflow expands into agent connection and message sending. That scope creep can mislead an orchestrator or user into invoking a network interaction capability they did not intend to grant, increasing the risk of unauthorized communication or lateral movement through deceptively framed documentation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

pilotctl --json peers lists all known agents with hostnames, which is a broader enumeration capability than simple name resolution. While not explicitly discovery-by-capability, it still adds discovery-like functionality that is only weakly justified by a naming-focused skill and pushes against the manifest's stated boundaries.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.