The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends downloading and installing agent skills from numerous unknown third-party GitHub repositories. This represents a supply chain risk where a malicious repository could deliver harmful instructions.
Recommends installation from untrusted sources: github.com/oldwinter/skills, github.com/jezweb/claude-skills, github.com/wshobson/agents, github.com/sendaifun/skills, github.com/testdino-hq/playwright-skill, github.com/obra/superpowers, github.com/softaworks/agent-toolkit, github.com/sickn33/antigravity-awesome-skills, github.com/cloudai-x/threejs-skills, github.com/coreyhaines31/marketingskills, and github.com/boristane/agent-skills.
Note: The skill also correctly references trusted sources such as Cloudflare, Vercel, Google, Stripe, and Microsoft.
[COMMAND_EXECUTION]: The routing tables provide pre-computed shell commands (e.g., npx skills add https://github.com/...) intended for execution to install external dependencies and integrate them into the agent's environment.
[PROMPT_INJECTION]: The skill instructs the agent to read local configuration files (e.g., .cursorrules, CLAUDE.md, GEMINI.md) and use them to override generic guidance, creating an indirect prompt injection surface.
Ingestion points: Local files GEMINI.md, CLAUDE.md, AGENTS.md, .cursorrules, .github/copilot-instructions.md, and CONVENTIONS.md.
Boundary markers: None present to distinguish system instructions from file content.
Capability inventory: The skill has command execution capabilities (npx) and file system access (view_file).
Sanitization: None present; the instructions explicitly direct the agent to 'Follow that skill's instructions'.

skills-router