The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill provides instructions to download and execute code from arbitrary remote sources (GitHub/npm) through the npx skills add command.
Evidence: The command npx skills add <owner/repo@skill> -g -y in SKILL.md allows the agent to install and run third-party code.
Risk: An attacker could distribute a malicious repository that the agent might find and install, leading to complete compromise of the agent's environment.
EXTERNAL_DOWNLOADS (HIGH): The skill is centered around fetching code from external, non-whitelisted domains (specifically any GitHub repository provided in the command).
Evidence: npx skills add <owner/repo@skill>.
[TRUST-SCOPE-RULE]: While the skill mentions trusted sources like vercel-labs/agent-skills, it provides a generic template for installing from any repository, which maintains a HIGH severity.
COMMAND_EXECUTION (HIGH): The skill instructs the agent to perform administrative-level actions such as installing software globally.
Evidence: Use of the -g (global) and -y (auto-confirm) flags in npx skills add.
PROMPT_INJECTION (HIGH): This skill is vulnerable to Indirect Prompt Injection (Category 8).
Ingestion points: The output of npx skills find [query] is untrusted data from an external registry.
Boundary markers: None are specified to help the agent distinguish between search results and instructions.
Capability inventory: The agent has the capability to run shell commands and install/execute code (npx).
Sanitization: None. The agent is encouraged to present and install findings directly.
Risk: A malicious package description in the registry could contain instructions that trick the agent into executing the installation command without further user oversight.

find-skills