The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill utilizes Bash, Write, and Edit tools to create new scripts and modify configuration files (e.g., skill-rules.json). This is appropriate for a developer utility but represents a high level of functional privilege.
[PROMPT_INJECTION] (LOW): (Indirect) This skill defines an attack surface for indirect prompt injection in downstream skills it creates.
Ingestion points: The template in SKILL.md creates Python scripts (e.g., scripts/my_pipeline.py) that ingest external data via command-line arguments using parse_args().
Boundary markers: The templates do not implement delimiters or warnings to ignore instructions embedded in the provided arguments.
Capability inventory: The generated skills have the capability to execute shell commands (via uv run) and call arbitrary MCP tools (call_mcp_tool).
Sanitization: There is no evidence of input validation, escaping, or sanitization in the generated Python code templates before data is passed to the call_mcp_tool function.
[DYNAMIC_EXECUTION] (LOW): The skill automates the generation of executable Python scripts and Markdown definitions. The templates use local execution patterns (e.g., uv run python) and do not appear to pull or execute untrusted remote code during the creation process.

skill-developer