The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes shell scripts (scripts/link_audit.sh and scripts/ui-codex) that execute shell commands, Python code, and the codex CLI tool. These scripts are intended for maintenance tasks such as link verification and automated UI triaging.
[EXTERNAL_DOWNLOADS]: The scripts/link_audit.sh script performs network requests (HEAD and GET) using Python's urllib.request library. It scans the skill's reference files for URLs and attempts to verify their availability when the --allow-network flag is used.
[PROMPT_INJECTION]: The scripts/ui-codex script is vulnerable to indirect prompt injection because it accepts user input via the --issue argument and interpolates it directly into a complex prompt string passed to the codex exec command.
Ingestion points: User-supplied text from the --issue command-line argument in scripts/ui-codex.
Boundary markers: The script uses a heredoc structure but lacks explicit delimiters or specific instructions to the model to ignore embedded directives in the user-supplied issue text.
Capability inventory: The script invokes codex exec with workspace-write permissions, allowing the downstream agent to modify repository files.
Sanitization: No validation, escaping, or sanitization is performed on the user-provided string before it is used to construct the prompt.

frontend-ui-design