The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest user-supplied intent and convert it into standing instructions within CLAUDE.md and related files. This creates a surface for indirect prompt injection where malicious input could be interpreted as behavioral overrides for the agent.
Ingestion points: User input captured during the "Capture intent" phase of the workflow (SKILL.md).
Boundary markers: The skill recommends using HTML comments for maintainer notes, which are stripped before injection, but lacks explicit boundary markers for user-interpolated content.
Capability inventory: The skill facilitates file writing and reading across the project structure.
Sanitization: The skill relies on manual "Audit" and "Verification" steps by the agent/user rather than automated input filtering.
[DYNAMIC_CONTEXT_INJECTION]: The documentation and recipes within the skill mention the !command syntax used by some platforms to execute shell commands at load time. While the skill suggests benign uses like git status, this feature represents a capability for arbitrary command execution if misused in generated rule files.
[EXTERNAL_DOWNLOADS]: The skill contains multiple references to external documentation and community resources (e.g., code.claude.com, github.com/anthropics/claude-code). These are documented as sources for best practices and troubleshooting guidelines from established and well-known services.

claude-md-rules-creator