The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured JSON data through its validation script (validate-prompt.js). While this constitutes a data ingestion surface, the risk is negligible as the script uses strict Zod schema validation and does not feed the content into sensitive sinks like shell execution or network requests.
Ingestion points: scripts/validate-prompt.js reads user-generated or agent-generated prompt.json files.
Boundary markers: The skill uses JSON structure and Zod schemas to enforce data integrity.
Capability inventory: Limited to file reading and console output; no network access or destructive file writes were found.
Sanitization: Strict validation against predefined schemas in scripts/prompt-schema.js.
[COMMAND_EXECUTION]: The skill documentation correctly identifies the use of a local Node.js script for validation purposes. The script performs internal logic calculations (scoring and schema checking) and does not execute arbitrary shell commands.
[EXTERNAL_DOWNLOADS]: The package.json file lists zod as a dependency. This is a well-known, trusted library from the official NPM registry and poses no security risk.
[REMOTE_CODE_EXECUTION]: No patterns for remote code execution (e.g., curl | bash) or dynamic loading of untrusted code were detected. All scripts are local to the skill directory.

superimage-generator