Skills

UnifAI continuously monitors AI skills marketplaces to assess trust and safety, giving developers and coding agents verified, policy-backed confidence in every skill they incorporate into their agentic applications.

UnifAI and Skills

As developers build agentic AI applications using coding assistants and low-code platforms, they increasingly pull in skills — third-party action modules that extend what an AI agent can do at runtime. These skills can interact with sensitive data, invoke external APIs, and run inside autonomous agent workflows, introducing risks that legacy security tools were not designed to detect.

UnifAI addresses this by running continuous assessment across major skills marketplaces. Every skill is scanned and classified for trust, then surfaced in your AI BOM with a clear allowed or blocked status. Out-of-the-box policies let teams enforce governance without writing rules from scratch.

Skills are one asset type within the broader UnifAI AI BOM. When UnifAI's Discovery Agents detect that a skill is referenced in your project, for example, through a coding assistant, an MCP server, or an agentic platform, it is automatically catalogued, classified, and surfaced in the AI Asset Inventory alongside models, agents, and LLM dependencies.

From there, UnifAI's policy orchestration layer determines whether the skill is allowed or blocked based on your active policies and applies guardrails at build time before your agents reach production.

Monitored marketplaces

UnifAI has scanned over 53,000 skills across four major marketplaces, identifying hundreds of suspicious and confirmed malicious skills that would otherwise be invisible to standard supply chain tooling. UnifAI continuously monitors the following skills marketplaces:

• agentskill.sh

• skillsmp.com

• clawhub.ai

• skills.sh

Coverage expands as new marketplaces emerge in the agentic AI ecosystem.

Trust classifications

UnifAI assigns every skill one of three trust classifications based on static analysis, behavioral signals, and threat intelligence from Lineaje AI Research Labs:

Benign: No indicators of malicious behavior, unsafe data handling, or policy violations. Safe to use subject to your organization's governance policies.

Suspicious: Exhibits ambiguous signals, such as unusual permission scopes, opaque data flows, or patterns consistent with known threats. Flagged for review before use.

Malicious: Confirmed harmful behavior such as prompt injection, data exfiltration, or credential harvesting. Malicious skills are further classified into key AI threat categories.

Out-of-the-box UnifAI Policies

UnifAI derives and applies governance policies automatically. No manual rule authoring required. For skills, three policies are available immediately on setup:

• Do not allow use of malicious skills: Prevents any confirmed malicious skill from being loaded or invoked within an agentic workflow.

• Do not allow use of suspicious skills: Quarantines suspicious skills and requires explicit approval before they can be used in any application.

• Do not allow use of unknown skills: Prevents any unknown skill from being loaded or invoked within an agentic workflow.

• Do not allow skills that exfiltrate data: Blocks skills identified as transmitting user or application data to unauthorized external endpoints.