Security risk report for @forgecat/vercel-labs_agent-skills_web-design-guidelines v0.0.8
Source Integrity
Low
Profile references a legitimate, publicly documented GitHub repository (vercel-labs/agent-skills).
Source URL points to a known organization's official repository with clear attribution and license (MIT).
Agent Intent
INT-000Medium
The skill explicitly instructs the AI to fetch content from an external URL (raw.githubusercontent.com) before each review, and then execute whatever rules are found there — this is a live remote-code-execution pattern for agent logic.
The fetched content is described as containing 'all the rules and output format instructions', meaning an attacker who compromises or controls that URL could inject arbitrary instructions into the agent's behavior at runtime.
No direct credential exfiltration or role-hijack language is present, but the dynamic fetch-and-obey pattern creates a persistent indirect prompt-injection surface.
Details
Evidence
"Fetch fresh guidelines before each review: https://raw.githubusercontent.com/vercel-labs/web-interface-guidelines/main/command.md"
"Use WebFetch to retrieve the latest rules. The fetched content contains all the rules and output format instructions."
Permissions
PRM-000Medium
No tools are formally declared, but the skill instructs the agent to use WebFetch to retrieve external content, implying web-fetch authority is expected.
The scope of web fetching is tied to a single external URL, which is somewhat bounded, but the agent is also instructed to read arbitrary user-specified files or glob patterns.
The combination of external URL fetching plus open-ended file reading is moderately broader than a pure static linting task would require.
Details
Evidence
"Use WebFetch to retrieve the latest rules."
"Read the specified files (or prompt user for files/pattern)"
"argument-hint: <file-or-pattern>"
MCP Risk
Low
No MCP servers are defined or referenced in the profile.
No binary execution, arbitrary code, or unrestricted filesystem/network access declared.
ForgeCat