The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external CLI tools (ffprobe, ffmpeg) to extract metadata and generate test assets. This is performed using execFileSync, which passes arguments as a literal array to prevent shell injection. A regression test in scripts/lib/probe.test.mjs specifically verifies that filenames containing shell metacharacters do not lead to arbitrary command execution by attempting to break out of the command string.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources such as catalog.json and Markdown design files (frame.md, design.md). These inputs populate metadata fields like descriptions and brand tokens. The skill mitigates risks by using strict boundary checks (isWithin) to ensure all file-writing operations remain within the project root or the trusted global cache (~/.media). It also employs HTML escaping in its evaluation report generation to prevent cross-site scripting.
[SAFE]: The skill implements robust security engineering patterns, including mandatory SHA-256 hash verification for ingested assets, checks for symbolic and hard link escapes before performing file operations, and restricting all execution to local filesystem paths. No network acquisition or exfiltration logic is present in the codebase.

media-use