The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains a dynamic context injection command using the !command syntax. The command d=${CLAUDE_SKILL_DIR}; command -v skill-toc >/dev/null 2>&1 && exec skill-toc "$d"; s=$HOME/.claude/skills/plugin-utils/bin/skill-toc; [ -x "$s" ] && exec "$s" "$d"; ... is used to generate a table of contents at load time. It functions by checking for the presence of a local utility named skill-toc and executing it to list the skill's reference files. This behavior is localized to documentation generation and does not involve unauthorized data access or network activity.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied queries about legal citations (Ingestion point: User interaction). Boundary markers (present/absent): Absent. Capability inventory: The skill has no code for subprocess calls, file-writes, or network operations across all reference scripts (Capability inventory: None). Sanitization (present/absent): Absent.

bluebook