The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to manage its lifecycle and skill storage.
It uses mkdir -p and mv to create and archive skill directories in both project-local (.claude/skills/) and global (~/.claude/skills/) paths.
It performs file-writing operations to generate SKILL.md files, which define new executable agent capabilities.
[PROMPT_INJECTION]: The 'Mode 4: Auto-Create' feature introduces an indirect prompt injection vulnerability.
Ingestion points: Command names from user input or errors (e.g., 'Unknown skill: resonance') are parsed as intent hints.
Capability inventory: The skill is instructed to 'infer intent' and 'execute immediately' based on these names, involving potential shell commands or file writes.
Boundary markers: No explicit delimiters or safety instructions are provided to isolate the inferred intent from the command name.
Sanitization: There is no evidence of input validation or sanitization for the command names or inferred instructions before execution.

create-shortcut