The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation contains multiple code examples with hardcoded database credentials.
Evidence found in SKILL.md: db_url="postgresql://user:pass@localhost:5432/agno".
Evidence found in references/integration.md: db_url="postgresql+psycopg://ai:ai@localhost:5532/ai" and db_url="postgresql+psycopg://user:pass@localhost:5432/db".
Evidence of AWS account metadata leakage in references/agentos.md: load_balancer_certificate_arn="arn:aws:acm:us-east-1:497891874516:certificate/..." revealing account ID 497891874516.
[REMOTE_CODE_EXECUTION]: The documentation promotes and provides examples for executing remote code from unverified sources using package runners.
Examples in SKILL.md and references/agentos.md show the use of npx -y @openbnb/mcp-server-airbnb and uvx mcp-server-git to run Model Context Protocol (MCP) servers.
While some sources like Google and the Model Context Protocol organization are well-known, others like openbnb are third-party and unverifiable.
[COMMAND_EXECUTION]: The framework described in the documentation provides first-class support for tools that execute arbitrary shell commands on the host system.
The documentation in references/index.md and SKILL.md references the use of ShellTools and MCPTools which leverage subprocess execution to interact with local systems.
[NO_CODE]: The skill consists entirely of documentation and reference files (markdown) and does not contain any directly executable Python or JavaScript files.

agno