The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands and Node.js scripts to verify the environment and validate design outputs. This includes directory traversal using relative paths to check for files and execute utilities.
Evidence: ls ${CLAUDE_SKILL_DIR}/../../../tokens >/dev/null 2>&1 && echo KIT_OK || echo KIT_MISSING in SKILL.md.
Evidence: node ${CLAUDE_SKILL_DIR}/../../../scripts/accuracy_report.mjs and several other scripts in the 'Gate' section of SKILL.md.
[DYNAMIC_EXECUTION]: The skill relies on several external Node.js scripts to perform runtime audits and validation of the generated UI code.
Evidence: Use of gate.mjs, slop_tells.mjs, and taste_audit.mjs to evaluate files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user requests for UI designs while having access to command execution and the file system, creating a risk for injection if input is used to determine parameters for these commands.
Ingestion points: User requests such as "make me a website" or "build a settings screen" mentioned in the description of SKILL.md.
Boundary markers: Absent. There are no instructions to ignore embedded commands or use delimiters for user-provided data.
Capability inventory: Shell command execution via ls and node, and reading files from the parent file system.
Sanitization: Absent. The skill does not describe any validation or escaping of user input before it influences the agent's actions or script parameters.

design-screen