The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The script scripts/build_design_prompt.py generates a prompt for a downstream Claude CLI session by interpolating data from a user-supplied JSON file without sanitization or boundary markers.
Ingestion points: The main function in scripts/build_design_prompt.py reads an external file provided via the --idea-json argument.
Boundary markers: The build_prompt function places the title, description, and category fields into a Markdown list under an 'Idea Specification' header. It does not include explicit delimiters or instructions to the downstream agent to treat these fields as data rather than instructions.
Capability inventory: The generated prompt is intended to be piped into claude -p. Per the SKILL.md documentation, this downstream agent is granted powerful capabilities including Read, Edit, Write, Glob, and Grep tools, which could be abused if an injection succeeds.
Sanitization: There is no escaping, validation, or filtering of the input strings before they are formatted into the final prompt string.

skill-designer