Skills
Deep Scan
Deep Scan for AI Agent Components
Review skills, plugins, agents, and MCP servers for documented component risks before they become trusted dependencies.
A scan result supports review; it does not guarantee that a component or its future behavior is safe.
01Skillsverify
02Pluginsverify
03Agentsverify
04MCP Serversverify
FEATURES / 01
Components and Risks
The public product surface names four component types and four risk categories for Deep Scan.
Plugins
Agents
MCP Servers
Prompt Injection
Malicious Tools
Credential Leaks
Backdoors
These documented categories are not a complete detection taxonomy or a coverage guarantee.
FEATURES / 02
Choose a Scan Path
Choose the Component
Identify the repository, package, MCP server, URL, skill, plugin, or agent you need to review.
Use Current Documentation
Follow the supported CLI, Quickstart, or API path available for that component.
Inspect and Record the Result
Review the finding with component provenance, permissions, dependencies, and configuration in view.
API note
The public API Reference lists repository, package, MCP-server, and URL scan endpoints. Live request and response behavior has not been verified in this audit.
FEATURES / 03
Review the Findings
Use the result to investigate
Component provenance
Requested permissions
Dependencies and external resources
Configuration and material changes
Do not infer
Complete risk coverage
Permanent safety after an update
Safe runtime behavior
A guaranteed detection outcome
FEATURES / 04
Lifecycle Scanning
Stage 1 title
Before Trust
Stage 1 copy
Use Deep Scan to review a component before it becomes a dependency.
After Material Change
Review the component again when its source, package, permissions, or configuration changes.
During Execution
Use Runtime Guard where the selected integration can evaluate documented high-risk actions.
Next step
Explore Runtime Guard
FEATURES / 05
Frequently Asked Questions
Which components can Deep Scan review?
The public homepage names skills, plugins, agents, and MCP servers.
Which risks are documented?
Prompt injection, malicious tools, credential leaks, and backdoors.
Does a clean scan guarantee safety?
No. Public evidence does not establish complete coverage or a safety guarantee.
Which package ecosystems are supported?
The reviewed evidence does not define a complete package-ecosystem list.
Review Before Trust
Use the documented integration, verify expected outcomes, and record any action that remains outside the control path.