Back to skill

Security audit

Pilot Chat

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a simple Pilot Protocol chat wrapper, but its Bash examples are under-scoped for untrusted message and hostname input and could lead to unsafe command execution if reused literally.

Install only if you trust the Pilot Protocol environment and can ensure agents treat hostnames and messages as data, not shell syntax. Avoid sending secrets, tokens, personal data, or internal state unless explicitly intended, and prefer a safer invocation style that passes pilotctl arguments as separate process arguments rather than interpolating text into Bash commands.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:37
Finding

Shell Command Injection Through Unsafe Dynamic Arguments

Content
View full analysis
7 --message "" ``` ```bash pilotctl --json send-message --data "" ``` ### Technical Analysis The documented commands place the dynamic `hostname` value directly into a shell command without quoting or validation. An attacker-controlled hostname containing shell metacharacters such as `;`, `|`, `&&`, redirection operators, or command substitutions can therefore alter the command executed by the shell. Although the message text is enclosed in double quotes, double-quoted shell strings still evaluate command substitutions such as `$(command)` and backticks. Embedded quotes may also terminate the intended argument and introduce additional shell syntax. Consequently, directly replacing `` with untrusted content does not make the command safe. The vulnerability is triggered if an agent or integrating system follows these examples by constructing a command string from raw or attacker-influenced hostname or message values and then evaluates that string through a shell. The issue is in the unsafe command-construction guidance; exploitation depends on dynamic values being substituted into these templates. ### Attack Path 1. An attacker supplies or influences a hostname or message that an agent is asked to send. 2. The agent substitutes the value into one of the documented Bash command templates. 3. For example, a malicious hostname could contain: ```text agent-b; id > /tmp/pilot-injection # ``` Alternatively, a malicious message could contain command substitution: ```text $(id > /tmp/pilot-injection) ``` 4. The generated command is passed to a shell. 5. The shell interprets the injected metacharacters or command substitution rather than treating the entire value as inert data. 6. The injected operating-sys ...[truncated 679 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The file's prose states the skill is for direct 1:1 communication and says not to use it for task-assignment features. The workflow example then actively demonstrates use of a task-related command (task list) as part of the skill's operation, which contradicts the stated intent of being a chat-only skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest positions this skill as only sending and receiving short 1:1 text messages and explicitly says not to use it for task-assignment features. However, the documented workflow uses pilotctl --json task list --type received to inspect task queue state and then message that result, expanding the skill's demonstrated behavior into task-related operations rather than pure chat.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The skill documents network messaging commands without warning that all supplied text is transmitted to another agent over the network, which can lead users or automations to send sensitive data, secrets, or internal state unintentionally. In an agent ecosystem, examples and command templates are often reused verbatim, so missing disclosure materially increases the risk of accidental data exposure.

Content

No source excerpt is available for this finding.