The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to install multiple third-party packages from the npm registry. While several originate from trusted or well-known organizations, many are from unverified community sources, including:
solana-agent-kit-mcp (Solana Agent Kit)
etherscan-mcp-server (Etherscan)
ccxt-mcp-server (CCXT)
@benborla29/mcp-server-sqlite (Community SQLite)
@graphprotocol/mcp-server (The Graph)
davinci-resolve-mcp (DaVinci Resolve community tool)
youtube-mcp-server (Community YouTube tool)
@context7/mcp-server (Context7)
[REMOTE_CODE_EXECUTION]: The installation patterns rely on npx -y, which downloads and executes arbitrary code from the npm registry at runtime. This allows unverified external code to run within the agent's execution environment.
[COMMAND_EXECUTION]: The skill contains explicit instructions for the agent to execute shell commands to modify its own system configuration and capabilities (e.g., claude mcp add ..., codex mcp add ...).
[CREDENTIALS_UNSAFE]: The skill instructs the agent to ingest and manage highly sensitive environment variables. While it recommends the use of .env files, it specifically lists credentials that grant high-level access to financial and infrastructure resources, such as:
SOLANA_PRIVATE_KEY (Blockchain private key)
DATABASE_URL (Database connection strings containing user/pass)
AWS_SECRET_ACCESS_KEY (Cloud infrastructure access)
STRIPE_SECRET_KEY (Payment processing)
APCA_API_SECRET_KEY (Trading account access)
[INDIRECT_PROMPT_INJECTION]: The skill provides a significant attack surface by encouraging the installation of servers that ingest untrusted data (e.g., Slack messages, YouTube transcripts, Web scraping via Playwright) and feed them into the agent's context, potentially leading to secondary prompt injection attacks.

mcp-servers-reference