The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a video production pipeline using Node.js and shell scripts to automate rendering. The pipeline uses the npx remotion command to render video variants. A security issue is present in render-all.sh (referenced in references/batch-ad-pipeline.md), where user-controlled identifiers from a CSV file are used to construct shell commands without sanitization, creating a risk of command injection.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing untrusted data from an external CSV file.
Ingestion points: Data enters the system via variants.csv, specifically through fields such as id, hook, and cta which are processed by csv-to-props.js.
Boundary markers: No explicit sanitization or boundary markers are used to isolate the data within the processing logic or the generated JSON property files.
Capability inventory: The skill possesses the ability to write files to the local filesystem (fs.writeFileSync in csv-to-props.js) and execute shell commands (npx remotion in render-all.sh).
Sanitization: The csv-to-props.js script performs basic validation on color hex codes but lacks sanitization for the id field. This unsanitized input is used directly in file paths and shell command arguments, creating a vulnerability.
[COMMAND_EXECUTION]: The csv-to-props.js script uses the id field from the CSV to determine the output path for property files via path.join(outDir, "${row.id}.json"). Without proper path sanitization, this allows for directory traversal attacks, potentially allowing the skill to write files outside the intended directory.

ad-creative-video