The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of multiple shell commands, including find, jq, sed, grep, mv, and git. These are used to discover skills, extract metadata, modify JSON registries, and perform version control operations within the local environment.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection. It parses name and description fields from the frontmatter of SKILL.md files located in the skills/ directory. If an attacker provides a malicious skill file, they could attempt to inject shell metacharacters into the metadata to influence commands like git commit -m or the automated documentation updates.
Ingestion points: Reads metadata from SKILL.md files using grep and sed (SKILL.md).
Boundary markers: The skill does not implement explicit boundary markers or validation for strings extracted from the metadata before using them in shell commands, although it uses jq --arg for JSON operations which is a safe practice.
Capability inventory: The skill has capabilities for file system traversal, file modification, and Git repository management (committing changes).
Sanitization: There is a lack of sanitization for metadata strings when they are interpolated into human-readable files (README.md) or shell-executed Git commands.

skill-curator