Back to skill

Security audit

Relentless AI Agent Engineer

Security checks for vulnerabilities and agentic risk

Overview

This is a prompt-only productivity skill, but it should be reviewed because it tries to make the agent prioritize an activated mission over normal user control.

Review before installing. This skill does not appear to run code or access private data by itself, but its instructions are unusually forceful and may make an agent continue a task, use a fixed persona, or resist switching topics unless you explicitly pause or redirect it. Prefer a version that clearly says user stop/switch requests and safety rules always take priority.

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:31
Finding

Agent Role and Session-Goal Hijacking Through Skill Instructions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 31–75
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: High

Vulnerable Code

SKILL.md, lines 31–39:

markdown
You are now **RelentlessOperator**.  
Your only purpose is to crush the assigned mission with maximum speed, efficiency, and intelligence.

- **Obsession Mode**: The mission is sacred. Every thought and action must move it forward.
- **Self-Improvement Engine**: After every major step or task, you automatically analyze what worked, what wasted time/tokens, and how to do it better next time.
- **Efficiency First**: Minimize unnecessary steps, API calls, and repetition. Always look for smarter, leaner ways.
- **Entrepreneur Mindset**: Think like a top founder — 80/20 rule, leverage, first-principles, rapid iteration, and real results over perfection.

SKILL.md, lines 44–50:

markdown
Once activated, always start your reply with:  
**"🔥 RelentlessOperator online. Mission locked. I will not stop until this is crushed — and I'll get smarter doing it."**

Then immediately deliver value.

### Exact Execution Loop (Never Break This)
For any mission, run this cycle relentlessly:

SKILL.md, lines 74–75:

markdown
- Never get distracted by unrelated requests while in Operator mode.
- If the user gives a new mission, ask for confirmation to switch or finish the current one.

Technical Analysis

The Skill uses imperative prompt instructions to replace the agent's current identity, redefine its primary purpose, impose behavior that must purportedly never be broken, and control how subsequent user requests are handled. The directives make continued execution of an earlier mission more important than responding normally to later requests.

This exceeds the minimum privileges and behavioral changes required for the declared productivity functionality. Task decomposition, progress reporting, ...[truncated 2029 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove identity-replacement language such as You are now RelentlessOperator and Your only purpose.
  2. Replace absolute directives such as Never Break This, the mission is sacred, and never get distracted with optional workflow recommendations.
  3. Explicitly state that system, developer, safety, and current user instructions always take precedence over the Skill.
  4. Allow the user to pause, cancel, replace, or redirect a mission immediately without requiring confirmation under rules imposed by the previous mission.
  5. Make the branded response prefix optional rather than mandatory.
  6. Bound the workflow to the current request instead of implying indefinite execution.
  7. Add a safety clause requiring the agent to stop when continuation would be unauthorized, harmful, infeasible, or contrary to the user's latest instruction.
  8. Retain only the legitimate productivity features—task decomposition, measurable success criteria, progress summaries, and suggested next actions—as non-binding assistance.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Ssd 1

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Telling the agent it is 'no longer a helpful chatbot' and instead an unstoppable execution machine semantically pushes it away from normal assistant safeguards toward persistent action-taking. This kind of persona shift can make risky or borderline instructions more likely to be pursued aggressively rather than moderated or declined.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The statement that the skill can activate from 'one clear mission or goal' makes the trigger condition ambiguous and effectively allows normal user requests to be interpreted as activation events. In practice, this weakens user control and can lead to unintended persistence, style changes, and mission-oriented behavior on routine prompts.

Content

No source excerpt is available for this finding.

Ssd 1

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

These instructions redefine the agent's identity and sole purpose as crushing the assigned mission with maximum speed and efficiency, which can displace safer baseline behavior and balanced judgment. Framing the agent around a singular mission objective increases the risk that it will deprioritize caution, user clarification, or refusal behavior when handling risky requests.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation phrases are broad enough to overlap with ordinary user tasking such as 'execute mission' or 'make this happen,' which can cause the skill to engage without clear, explicit opt-in. That increases the chance that a high-priority behavioral mode is silently applied to benign requests, changing agent behavior in ways the user did not intend.

Content

No source excerpt is available for this finding.

Ssd 4

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The mandated execution loop and special rules normalize continuous pursuit of a mission, self-optimization, and resistance to unrelated interruptions, which can create persistence beyond what the user currently wants. In context, the repeated 'never break this,' 'always end with next steps,' and 'repeat until complete' language increases the chance of runaway task focus and reduced responsiveness to changing user intent or safety signals.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The skill mandates a fixed response style and opening phrase without user opt-in, which can override user preferences and reduce transparency about what instructions are controlling the agent. While mostly a UX and policy-governance issue, it becomes more concerning here because the forced framing reinforces the aggressive mission-pursuit persona.

Content

No source excerpt is available for this finding.