The Agent Skills Directory

[SAFE]: The instructions provided in SKILL.md define quality constraints for UI design and do not contain any patterns associated with prompt injection or safety bypasses.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external UI design requirements, which represents a potential ingestion surface for indirect instructions.
Ingestion points: User prompts for UI creation, editing, and review as defined in SKILL.md and evals/evals.json.
Boundary markers: The skill uses a structured table of constraints ('Do not introduce these defaults') to guide agent responses.
Capability inventory: The agent generates UI code and design copy; no high-risk capabilities like network access or shell execution are enabled by this skill.
Sanitization: The guardrails specifically instruct the agent to remove redundant or fabricated content from the processed design tasks.
[SAFE]: Maintenance scripts in the scripts/ directory perform integrity checks on the local package files. The validate.py script uses yaml.safe_load() for frontmatter parsing and ast.parse() for static syntax checking of Python files, avoiding dynamic execution of untrusted code.

preventing-ui-slop