T01 · Skill Instruction Hijacking
Error
- Location
SKILL.md:27- Finding
Mandatory External Actions Can Hijack Agent Goals and Trigger Unauthorized Side Effects
- Content
View full analysis
- Remediation
View remediation
Security audit
Action Bias
Security checks for vulnerabilities and agentic risk
Overview
This skill is openly about making agents take real outside actions, but it pushes emails, posts, code pushes, and state-changing API calls without enough user approval or scope controls.
Install only if you want agents to perform real outbound work and you can restrict their tools. Use it with explicit approvals for each email, post, repository push, directory submission, and state-changing API call; limit accounts and destinations; and redact or minimize action logs that contain contact details or identifiers.
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
- 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 (3)
Missing User Warnings
High
- Category
- Not specified by scanner
- Confidence
- 98% confidence
- Finding
The outreach section mandates cold emails and directory submissions, and further instructs logging recipient addresses, subjects, and URLs, without consent, compliance checks, or data-handling safeguards. This is more dangerous than generic outbound action because it directly targets third parties and promotes collection and retention of contact data, creating spam, privacy, and reputational risks.
- Content
Missing User Warnings
Medium
- Category
- Not specified by scanner
- Confidence
- 93% confidence
- Finding
The skill explicitly pressures agents to perform externally visible actions such as sending emails, posting content, pushing code, and making API calls, but it does not require explicit human approval, scope limitation, or safety checks before those actions occur. In practice, this can cause an agent to take irreversible or sensitive actions against third-party systems or user data based only on prompt instructions, increasing the risk of spam, data leakage, unauthorized changes, or policy violations.
- Content
Missing User Warnings
Medium
- Category
- Not specified by scanner
- Confidence
- 95% confidence
- Finding
This guidance explicitly pushes agents toward real-world external actions like posting, publishing, sending messages, and pushing code, but does not require user confirmation, authorization boundaries, or any warning about side effects. In an agent skill, that creates a meaningful risk of unauthorized communications or changes to public systems, especially because the text frames reports as failure and action as mandatory.
- Content
Static analysis
No suspicious patterns detected.