Security audit

Context Mode

Security checks for vulnerabilities and agentic risk

Overview

This skill is a research helper that uses fetching, indexing, and command-style analysis tools in ways that mostly match its stated purpose, with some scope and consent caveats.

Install this only if you want an agent to use context-mode tools for research, indexing, and scripted analysis. Do not use it for sensitive documents unless you are comfortable with indexing and later reuse, and prefer explicit user approval before direct shell commands or curl-based fetching.

A.I.G

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

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Vague Triggers

High

Confidence: 97% confidence
Finding: The trigger phrases are extremely broad and map to common requests like 'check this,' 'analyze this content,' or 'latest news,' causing the skill to activate in many ordinary conversations without clear user consent. Overbroad activation is especially risky here because the skill then steers the agent toward indexing, searching, and execution-capable tools, potentially changing the agent's behavior in security-relevant ways unexpectedly.

Natural-Language Policy Violations

Medium

Confidence: 87% confidence
Finding: Mandating a fixed persona and style without user opt-in can override user preferences and system behavior, reducing transparency about when the skill is influencing outputs. While not directly enabling code execution, this kind of hidden steering can make unsafe tool usage or workflow changes less visible to the user in a skill that already pushes powerful research and execution primitives.

Context-Inappropriate Capability

Medium

Confidence: 90% confidence
Finding: The skill explicitly promotes generic execution tools (`ctx_batch_execute`, `ctx_execute`) with shell/script examples that go beyond a narrowly bounded research workflow. In an agent setting, encouraging broad execution increases the chance the model will run arbitrary commands on local data or tooling surfaces when a user only asked for content research, expanding attack surface and enabling command misuse.

Context-Inappropriate Capability

Medium

Confidence: 94% confidence
Finding: The multi-source example tells the agent to use raw `curl` commands to fetch external sources, bypassing the skill's safer bounded fetch-and-index pattern. This is dangerous because it normalizes unrestricted network command execution, which can be repurposed for unauthorized outbound access, SSRF-style behavior through available network paths, or retrieval of untrusted content via shell commands.

Static analysis

No suspicious patterns detected.