The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input from Q&A sessions to generate implementation plans, which is a standard functional requirement but creates a minor injection surface.
Ingestion points: User clarifications from the blueprint Q&A conversation are synthesized into prompts and plan files via references/refine-prompt.md.
Boundary markers: There are no explicit delimiters used to isolate user input during interpolation, although the skill enforces a strict bulleted structure for synthesis.
Capability inventory: The skill is capable of directory creation, codebase exploration, and writing markdown files.
Sanitization: Slugs for file paths are sanitized for safety, though the synthesized plan text is not specifically filtered for malicious instructions.
[COMMAND_EXECUTION]: The skill uses shell commands to organize its output.
Evidence: SKILL.md invokes mkdir -p to create the implementation plan directory.
Mitigation: The skill sanitizes the slug variable to lowercase alphanumeric characters and hyphens, preventing command injection via path manipulation.

blueprint-generate