The Agent Skills Directory

[SAFE]: The skill provides structured guidance and templates for creating AI agent skills within the ecosystem. It explicitly documents security risks like prompt injection via XML tags in frontmatter and provides remediation guidance.
[DYNAMIC_EXECUTION]: The skill references the execution of local validation scripts (e.g., scripts/lint-skills.sh, scripts/scan-skills.sh). These are part of the intended developer workflow for quality assurance and security auditing of new skills. The skill itself restricts tool access to filesystem operations (Read, Write, Edit, Glob, Grep), effectively preventing it from executing these shell scripts directly unless the agent is granted broader permissions in a different context.
[INDIRECT_PROMPT_INJECTION]: As a document generator, the skill has an attack surface for indirect prompt injection. It ingests user prompts to create instructions for agents. The skill mitigates this by using YAML frontmatter delimiters and explicitly warning against the inclusion of XML tags that could influence the underlying model's system prompt. Capability inventory is limited to filesystem tools, and sanitization guidelines are provided.

skill-writer