The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill explicitly directs the agent to install and run additional skills from an untrusted third-party source (github.com/Leonxlnx/taste-skill) using the 'npx skills add' command for tasks such as UI redesign and image-to-code generation.
[COMMAND_EXECUTION]: The 'Site readiness check' automatically executes shell commands to move files, modify repository configuration (e.g., .gitignore), and alter CI/CD workflows (.github/workflows/deploy.yml). It also relies heavily on 'php forge', 'composer', and 'npm' for system-level operations.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads project-specific context and instructions from 'docs/ai/PRODUCT.md'.
Boundary markers: There are no delimiters or warnings to ignore embedded instructions when parsing these files.
Capability inventory: The skill has authority to execute shell commands, install packages, and manage credentials via forge.
Sanitization: Content from 'PRODUCT.md' is used to influence agent behavior and parameterize CLI commands without validation or escaping.
[CREDENTIALS_UNSAFE]: The skill facilitates the retrieval of secrets from a Bitwarden vault using 'php forge credentials' and writes them to a local '.env' file. While following standard secret management patterns for storage, the automated extraction of credentials from a remote provider is a high-risk operation.
[EXTERNAL_DOWNLOADS]: The skill downloads project manifests from 'github.com/wonder-image/lib' and installs development tools from the official 'playwright' package and 'vercel-labs' (a trusted organization).

wi-site