The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides installation commands for official Anthropic SDKs across several ecosystems (npm, PyPI, Maven, Gradle, NuGet, Composer, Gem, and Go). It also includes instructions for downloading the 'ant' CLI from the official Anthropic GitHub repository.
[PRIVILEGE_ESCALATION]: Installation instructions for the 'ant' CLI on Linux systems include a command that uses sudo to extract the binary to the /usr/local/bin directory. This is a standard system-level installation practice for command-line tools to ensure they are available in the system's execution path.
[INDIRECT_PROMPT_INJECTION]: The skill outlines capabilities for agents to ingest data from untrusted external sources, such as through web search, web fetch, and reading files from GitHub repositories or user uploads. These data ingestion points are used in conjunction with powerful capabilities like shell command execution (bash) and server-side code execution. To mitigate the associated risks of indirect prompt injection, the skill includes extensive documentation on security measures, including sandboxing, network egress restrictions, and guidance on auditing prompts for dated or risky patterns.

claude-api