Back to skill

Security audit

IAJUS — Verificar Citações BR

Security checks for vulnerabilities and agentic risk

Overview

This skill provides proportionate citation checks through a disclosed external service, with privacy considerations and some broader research examples in its documentation.

Use public citations and avoid confidential client details in queries. IAJUS states that filtered query terms are retained for 90 days and semantic queries reach AI providers; review its [privacy policy](https://iajus.com.br/privacidade). Review OAuth consent and service pricing, keep searches tied to supplied citations, and check returned official sources before relying on results.

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
Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

The workflow sends quoted text to an external service without warning about disclosure of potentially confidential material.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The workflow places a user-supplied quotation in a query sent to the hosted IAJUS service without first checking whether it contains confidential or restricted information. Although the intended inputs are quotations from public judicial decisions and the service endpoint is disclosed, an unverified quotation could contain private case details or client material, creating an unintended third-party disclosure. Restricting requests to targeted excerpts reduces exposure but does not establish authorization to disclose sensitive content.

Content

No source excerpt is available for this finding.

The skill instructs the agent to conduct open-ended jurisprudence research despite the manifest explicitly excluding research from scratch.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest limits the skill to checking supplied citations for existence, fidelity, and current validity and explicitly says not to research from scratch. L009–L030 instead prescribe topic-based searches, pagination, and finding support for assertions without requiring an existing citation; L104 further directs research in response to general questions about rights, duties, or legal regimes. These instructions support discovering legal authorities rather than merely locating and verifying cited ones.

Content

No source excerpt is available for this finding.

The reference directs jurisprudence research beyond verification of supplied citations.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The manifest limits the skill to verifying existing citations and explicitly excludes researching from scratch. These instructions instead prescribe ontology-filtered decision searches and, at L09, hybrid or semantic searches for broad concepts without requiring an existing citation to verify.

Content

No source excerpt is available for this finding.