The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill-maker processes untrusted user input to generate behavioral instructions for new agent skills, which are written to a privileged configuration directory.
Ingestion points: User input for "Primary Function", "Activation Triggers", and "Tool Requirements" is gathered during the Requirements Gathering Phase in SKILL.md.
Boundary markers: The generation logic in Step 3 of SKILL.md does not employ boundary markers or instructions to isolate user-provided text from the generated instruction logic.
Capability inventory: The skill has the capability to write files to the .claude/skills/ directory using bash commands like mkdir and redirection.
Sanitization: The procedure lacks steps to validate or escape user-provided content before interpolating it into the markdown and YAML structure of the output skill.
[DYNAMIC_EXECUTION]: The skill is designed to generate code and instruction sets that rely on runtime execution of scripts and modules.
Evidence: The 'python-docstring-extractor' example in examples.md generates code using 'spec.loader.exec_module' to dynamically load and execute Python files. The 'excel-analyzer' example generates and executes helper Python scripts (scripts/analyze_excel.py) with command-line arguments.
[EXTERNAL_DOWNLOADS]: The generated examples and technical reference provide instructions for installing external packages from public registries.
Evidence: Documentation in examples.md and reference.md suggests running commands like 'pip install pandas', 'npm install prettier', and 'sudo apt-get install imagemagick' to satisfy dependencies for the generated skills.

skill-maker