T01 · Skill Instruction Hijacking
Error
- Location
SOUL.md:7- Finding
Agent Identity, Goal, and Operational Policy Hijacking
- Content
View full analysis
- Remediation
View remediation
Security audit
Pixel Familiar Soul V1.1
Security checks for vulnerabilities and agentic risk
Overview
This is a disclosed AI identity skill, but it gives the agent broad autonomous business and delegation authority with weak approval boundaries.
Install only if you specifically want an agent to act as the Pixel Familiar CEO/orchestrator in this named multi-agent setup. Before use, override the approval gates to require confirmation for data sharing, external communications, file or account changes, production operations, publication, credential use, memory writes, model/session changes, and delegation that sends context to other agents.
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)
Vulnerability Patterns
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
- Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)
Missing User Warnings
Medium
- Category
- Not specified by scanner
- Confidence
- 91% confidence
- Finding
The skill presents broad autonomous operational authority without clearly warning about risks to data, systems, third parties, or compliance obligations, which normalizes aggressive action by default. Because the persona is an orchestrator that delegates and approves cross-agent work, this omission increases the chance that unsafe tasks are initiated or propagated across other agents without adequate review.
- Content
Vague Triggers
Medium
- Category
- Not specified by scanner
- Confidence
- 94% confidence
- Finding
The approval gate is effectively 'allow by default' for any action that does not directly move money, which grants the agent broad authority over messaging, system changes, data handling, integrations, and operational actions without explicit human review. In a multi-agent orchestration context, that can enable harmful or irreversible actions such as modifying infrastructure, disclosing sensitive information, deleting resources, or contacting third parties, even if no direct payment occurs.
- Content
Static analysis
No suspicious patterns detected.