Back to skill

Security audit

wikitolica-search

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent Wikitólica lookup helper, but it asks for unrestricted shell access even though its task only needs scoped web retrieval.

Review this skill before installing because it can grant the agent unrestricted shell access whenever the skill is active. The content workflow itself is straightforward, but a safer version would remove Bash(*) and use only scoped fetch or MCP tools for Wikitólica pages.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (1)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:14
Finding

Unrestricted Shell Permission Exceeds Functional Requirements

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 14
Vulnerability Type: Excessive tool permissions and violation of least privilege
Risk Level: Medium

Complete Code Snippet:

yaml
allowed-tools: Bash(*) Fetch(*)

Technical Analysis

The Bash(*) declaration permits unrestricted shell command execution. The documented workflow only needs to retrieve the Wikitólica sitemap and article pages over HTTPS, as shown by the direct HTTP method in lines 32–37. Those operations can be implemented with a narrowly scoped fetch capability, so arbitrary shell access is unnecessary.

Unrestricted Bash access expands the Skill's authority from web retrieval to the full permissions of the Agent process. Commands could inspect or modify local files, invoke installed programs, access environment variables, and connect to unrelated resources. Although the audited Skill does not explicitly instruct the Agent to misuse Bash, the broad permission creates a dangerous execution channel if the Agent is influenced by malicious user input or adversarial remote content.

Attack Path

  1. The Skill is activated for a Catholic-related query.
  2. The Agent loads the Skill and receives unrestricted Bash(*) access.
  3. An attacker supplies crafted input, or compromised remote content contains instructions that influence subsequent Agent behavior.
  4. The Agent interprets that content as a reason to invoke Bash rather than the dedicated fetch capability.
  5. An arbitrary command executes with the operating-system privileges and resource access of the Agent process.
  6. Depending on the runtime environment, the command may read or alter accessible files, expose environment data, launch other programs, or communicate with unrelated network destinations.

Impact Assessment

Successful exploitation could provide command execution under the Agent process's account. The precise scope depends on sandboxing and host permissions, ...[truncated 518 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove Bash(*) and use a dedicated HTTP retrieval tool for the documented sitemap and article requests.
  2. Restrict network access to HTTPS requests whose normalized hostname is exactly www.wikitolica.com.
  3. Permit only the required paths, such as /sitemap.xml and validated article paths matching /{initial}/{slug}/.
  4. Reject redirects to other hosts and validate the final destination after every redirect.
  5. Treat fetched content as untrusted data and prohibit it from authorizing tool calls or changing Agent instructions.
  6. If shell access is operationally unavoidable, replace Bash(*) with a narrowly allowlisted command wrapper that accepts no shell metacharacters, validates all URL components, and invokes the HTTP client without shell interpolation.
  7. Apply runtime sandboxing, filesystem restrictions, environment-variable isolation, and outbound network controls as defense in depth.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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 (2)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

La activación para términos 'identificables como católicos o relacionados con el Catolicismo' es demasiado amplia y subjetiva, lo que puede disparar la skill en consultas tangenciales o ambiguas. Esto puede secuestrar el enrutamiento de la conversación, introducir una fuente especializada cuando no corresponde y degradar la precisión o neutralidad de la respuesta.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

La regla 'responde en español' impone un idioma fijo independientemente del idioma del usuario o sus preferencias. Esto puede causar respuestas no solicitadas, mala experiencia de usuario y errores de cumplimiento con expectativas del sistema o del usuario, aunque no supone por sí mismo una explotación técnica severa.

Content

No source excerpt is available for this finding.