The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include multiple shell command blocks (ffmpeg, ffprobe, file, du) that incorporate shell variables $INPUT_FILE and $OUTPUT_FILE. If a user provides a filename containing shell metacharacters (e.g., ; rm -rf /), it could lead to arbitrary command execution during the processing phase.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes external files and their metadata.
Ingestion points: The skill ingests untrusted data via the $INPUT_FILE variable in SKILL.md (Step 1 and Step 3).
Boundary markers: None are defined to isolate the untrusted input variables from the shell commands.
Capability inventory: The skill uses ffmpeg, ffprobe, file, du, cut, and head (SKILL.md).
Sanitization: There is no evidence of filename sanitization or validation of input content beyond checking for file existence.

gif-generation