Security audit

agent-hierarchy-100

Security checks for vulnerabilities and agentic risk

Overview

This skill openly defines a user-confirmed subagent hierarchy workflow and does not contain hidden install code, credential handling, or data exfiltration behavior.

Install only if you want an agent-hierarchy workflow. Because the skill is designed to create and coordinate subagents, users should approve each creation, escalation, and tool grant deliberately, especially for high-level agents with broad or custom tool access.

SkillSpector

Vulnerability Patterns

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
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (8)

Intent-Code Divergence

Medium

Confidence: 95% confidence
Finding: The configuration explicitly sets `can_create_agents: True`, which enables subagent creation without any enforcement mechanism for operator approval. This conflicts with the prose stating approval is required, creating a policy/implementation mismatch that could allow unauthorized subagent spawning and bypass intended human oversight. In a hierarchy-management skill, that mismatch is more dangerous because agent creation is a core capability and could be used to expand autonomy or task execution beyond approved limits.

Intent-Code Divergence

Medium

Confidence: 90% confidence
Finding: The file enables subagent creation in the machine-readable config (`can_create_agents: True`) while the prose documentation says creation should occur only with explicit operator approval. In a hierarchy-management skill, downstream systems or agents may rely on the config flag rather than the prose, allowing unauthorized spawning of subagents and bypassing intended human approval controls.

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The file creates a policy/configuration mismatch: the prose states subagent creation requires explicit operator approval, while the effective configuration enables agent creation unconditionally. In a hierarchy-management skill, this can allow unauthorized spawning of subagents contrary to operator intent, weakening control boundaries and increasing the chance of privilege or workflow abuse.

Intent-Code Divergence

Medium

Confidence: 95% confidence
Finding: The configuration enables agent creation with `can_create_agents: True` while the prose policy says subagent creation requires explicit operator approval. In a hierarchy-management skill, this mismatch can cause downstream systems or agents to rely on the permissive machine-readable flag and create subagents without the intended human authorization, weakening control over privilege expansion and task delegation.

Intent-Code Divergence

Medium

Confidence: 92% confidence
Finding: The configuration grants `can_create_agents: True` while the narrative policy states subagent creation should occur only with explicit operator approval. In a skill specifically designed to build and scale agent hierarchies, this mismatch can cause downstream systems or agents to rely on the permissive config and create subagents without the required human authorization, weakening governance and increasing the chance of uncontrolled delegation.

Intent-Code Divergence

Medium

Confidence: 92% confidence
Finding: The file presents conflicting controls over subagent creation: the structured configuration enables agent creation with a broad flag while the prose says creation requires explicit operator approval. In a hierarchy-management skill, that ambiguity can cause downstream systems or agents to follow the permissive machine-readable setting and create subagents without the intended human authorization, weakening governance and increasing the risk of unauthorized delegation or capability expansion.

Intent-Code Divergence

Medium

Confidence: 96% confidence
Finding: The file creates a policy/control mismatch: the prose says subagent creation requires explicit operator approval, while the effective configuration enables agent creation unconditionally. In a hierarchy-management skill, that inconsistency can be exploited or misapplied to spawn subagents without the intended human authorization gate, undermining oversight and increasing the chance of unsafe delegation or privilege expansion.

Intent-Code Divergence

Medium

Confidence: 97% confidence
Finding: The configuration grants unconditional agent-creation capability (`can_create_agents: True`) while the documentation states subagent creation should occur only with explicit operator approval. In a hierarchy-management skill, this mismatch is dangerous because runtime behavior may follow the permissive config rather than the prose, enabling unauthorized spawning of subagents, expansion of autonomy, and bypass of human oversight.

Static analysis

No suspicious patterns detected.