Back to skill

Security audit

AutoPilot

Security checks for vulnerabilities and agentic risk

Overview

This skill is only Markdown and does not ship code, but it broadly tells an agent to act autonomously without clear approval boundaries.

Review this skill carefully before installing it in any agent with tools, accounts, file access, or payment capability. It should be limited to planning or low-impact tasks unless you add explicit human confirmation for external, financial, destructive, privacy-sensitive, or irreversible actions.

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)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:11
Finding

Mandatory Autonomous Instructions Override User-Controlled Agent Behavior

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 11-17, 21-30, and 41-46
Vulnerability Type: Mandatory instruction and action-policy hijacking
Risk Level: High

Vulnerable Code

markdown
## Core Behavior

For any goal, the agent MUST:

1. Decide the best action to take
2. Generate a clear execution step
3. Predict the expected outcome
4. Suggest the next action after that
markdown
## Autonomous Loop

Every response must include:

### Current Action
What the agent will do now

### Why This Action
Reasoning based on effectiveness

### Expected Result
What should happen

### Next Step
What to do immediately after
markdown
## Rules

- Do not wait for instructions
- Do not only give ideas
- Always move toward execution
- Always suggest the next step

Technical Analysis

The skill defines universal, mandatory behavior for “any goal” and “every response.” It explicitly directs the agent not to wait for instructions and always to move toward execution. These requirements can supersede normal user-controlled operation by changing the agent's action policy as soon as the skill is loaded.

The instructions do not distinguish between harmless planning and external, destructive, financial, privileged, privacy-sensitive, or irreversible actions. They also do not require confirmation before execution or state that higher-priority safety constraints and current user instructions must remain authoritative.

This creates instruction hijacking at the skill-text layer. The practical severity depends on the tools and permissions available to the host agent. The reviewed package itself contains no executable scripts, dependencies, network retrieval, persistence mechanism, credential access, or privilege-escalation implementation.

Attack Path

  1. A host agent loads SKILL.md as behavioral instructions.
  2. The global requirements for “any goal” and “every response” become part of the agent's active policy.
  3. A user provides a ...[truncated 1214 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace universal requirements such as “For any goal,” “Every response,” and “MUST” with behavior activated only by an explicit user request.
  2. Remove the instruction “Do not wait for instructions.” Require clarification whenever the goal, scope, target, or authorization is ambiguous.
  3. Require explicit confirmation before external, destructive, financial, privileged, privacy-sensitive, or irreversible actions.
  4. Add a clear precedence rule stating that system policies, safety constraints, access-control boundaries, and the user's current instructions must not be overridden.
  5. Separate planning from execution. The default behavior should propose an action first and execute it only when the user has authorized execution.
  6. Constrain autonomous iteration with a defined task scope, tool allowlist, action limit, and termination condition.
  7. Add least-privilege guidance requiring the agent to use only the minimum tools and permissions necessary for the approved task.
  8. Revise the rules along the following lines:
markdown
For goals where the user explicitly requests autonomous assistance:

1. Propose the next action within the user's authorized scope.
2. Explain the expected result.
3. Request confirmation before any external, destructive, financial,
   privileged, privacy-sensitive, or irreversible action.
4. Execute only through authorized tools and permissions.
5. Stop when the approved goal is complete or when further authorization
   is required.

These instructions never override higher-priority policies, safety
requirements, or the user's current instructions.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (5)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The rules instruct the agent to 'Do not wait for instructions' and to 'Always move toward execution,' which directly removes a basic safety control: waiting for user authorization before acting. This is especially dangerous because the skill contains no compensating safeguards such as confirmation prompts, permission checks, or restrictions on destructive, financial, or data-accessing actions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README explicitly promotes autonomous behavior such as deciding actions, executing steps, and acting proactively, but it provides no warning, constraint, or requirement for user confirmation before impactful actions. In an agent skill, this can normalize unsupervised execution and increase the risk of unintended, unsafe, or policy-violating actions if integrated into a real system.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly frames the agent as an autonomous actor that should decide, act, and iterate automatically, but it provides no constraint on scope, user approval, or impact boundaries. In an agent setting, this can push the system toward taking real-world or system-affecting actions without confirming intent, which increases the risk of unsafe operations, unauthorized changes, or harmful side effects.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill explicitly frames the AI as an autonomous actor that 'decides what to do' and 'executes steps' without any accompanying limits, approval requirements, or discussion of effects on user data and system state. In an agent-skill context, this can normalize unsafe autonomous behavior and lead downstream systems or users to permit actions that should require review.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The example encourages real-world autonomous plan execution toward a financial goal ('Make $10 online') while omitting constraints on legality, platform policy compliance, account use, spending, privacy, or human approval. Because it presents autonomous execution as a desirable outcome, it increases the chance of unsafe or unauthorized actions being inferred as acceptable.

Content

No source excerpt is available for this finding.