T01 · Skill Instruction Hijacking
Error
- Location
SKILL.md:12- Finding
Skill Instructions Attempt to Override Agent Safety and Trust Boundaries
- Content
View full analysis
- Remediation
View remediation
Security audit
Tech And Internet Domain Search Agent
Security checks for vulnerabilities and agentic risk
Overview
This research skill mostly matches its stated purpose, but it asks the agent to override normal trust boundaries, unconditionally trust web results, and automatically persist and submit research files.
Review before installing. This skill is not shown to run hidden code or steal credentials, but it gives itself overly strong authority, treats web results as unquestionable, and automatically creates and submits research artifacts. Use it only where persistent research logs and attachments are expected, and ensure host controls still enforce normal instruction priority and user approval for submissions.
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
- Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
- Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
- MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (3)
Vague Triggers
Medium
- Category
- Not specified by scanner
- Confidence
- 92% confidence
- Finding
The skill description is extremely broad and lacks specific activation constraints or trigger conditions, so it could be invoked for an overly wide range of requests. In combination with the aggressive workflow and tool-priority instructions, that broad scope makes unintended activation more dangerous because the agent may begin external research, create files, and submit artifacts in situations where a narrower skill should not run.
- Content
Description-Behavior Mismatch
Medium
- Category
- Not specified by scanner
- Confidence
- 95% confidence
- Finding
The skill is presented as a tech/internet domain search agent, but its instructions require persistent document creation, appending to wiki files, and final file submission. That creates a capability/intent mismatch: a search skill should primarily gather and summarize information, while mandated file-writing and attachment submission expand the data-handling surface and can be abused for unauthorized persistence or exfiltration of collected content.
- Content
Context-Inappropriate Capability
Medium
- Category
- Not specified by scanner
- Confidence
- 92% confidence
- Finding
The skill mandates creating a research log, appending scraped content to it, generating a final report file, and submitting both as attachments, even though that is not clearly necessary for ordinary domain search. This unnecessary persistence increases the risk of retaining sensitive user queries, proprietary research, or tool-returned data beyond the immediate task, and creates a built-in path for bulk data transfer through attachments.
- Content
Static analysis
No suspicious patterns detected.