The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/frame.sh script constructs a complex ffmpeg filter string by directly interpolating the --index variable: -vf "select=eq(n\\,${index})". Because this input is not validated or sanitized, a malicious input could break out of the intended filter and inject arbitrary ffmpeg filters. This could potentially be used to read unauthorized local files using filters like subtitles or movie within the context of the ffmpeg process.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to exploitation via malicious data as it processes external files and allows specifying an arbitrary output path.
Ingestion points: Command-line arguments in scripts/frame.sh (input file, index, time, and output path).
Boundary markers: None present.
Capability inventory: Directory creation (mkdir -p), file system writes via ffmpeg, and shell command execution.
Sanitization: There is no validation of the format for the index or time parameters, nor are there restrictions on the destination of the --out path, allowing the agent to potentially overwrite sensitive files if directed by a malicious prompt.
[COMMAND_EXECUTION]: The skill executes shell commands (ffmpeg, mkdir, dirname) using variables derived from user/agent input. While the variables are quoted to prevent basic shell expansion, they remain vulnerable to option injection and the aforementioned filtergraph injection.
[EXTERNAL_DOWNLOADS]: The skill metadata specifies the installation of ffmpeg via the Homebrew package manager (brew). This involves fetching the formula and binary from official Homebrew repositories, which are well-known and trusted infrastructure.

video-frames