The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions in README.md and the install.sh file utilize the pattern of downloading a script via curl and piping it to bash. The source URL (raw.githubusercontent.com/marceloeatworld/mdbook-ai-skill/main/install.sh) is owned by the skill's author and is consistent with the vendor's own infrastructure. While this represents a remote code execution vector, it is a standard installation method for developer tools.
[EXTERNAL_DOWNLOADS]: The skill is configured to download reference documentation from the official mdBook GitHub repository (github.com/rust-lang/mdBook.git). It also includes instructions for the AI agent to fetch live updates from the official rust-lang GitHub organization when local references are insufficient. As these resources originate from a well-known and reputable technology organization, this is documented as normal behavior.
[COMMAND_EXECUTION]: The provided install.sh and scripts/generate-references.sh scripts perform local command-line operations including directory creation, file deletion, and git operations. These operations are scoped to the skill's own installation and reference directories.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes external markdown files from a well-known repository. 1. Ingestion points: Files in the references/ directory and raw content from the rust-lang GitHub organization. 2. Boundary markers: The SKILL.md file does not define explicit delimiters or warnings for instructions embedded within the fetched documentation. 3. Capability inventory: The skill provides shell scripts for installation and data generation, and the agent typically has file system access. 4. Sanitization: No content sanitization or validation is performed on the ingested markdown files before they are presented to the agent's context.

mdbook