ccr-import
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on local shell utilities including
adb,ffmpeg,awk, andgrepto interact with a connected Android device and manipulate audio files. These commands are necessary for the skill's stated purpose of call recording management. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing untrusted data (filenames) retrieved from an external mobile application.
- Ingestion points: External filenames are retrieved from the Android filesystem using the
adb shell lscommand inSKILL.md. - Boundary markers: Absent. There are no delimiters or instructions provided to the agent to treat filenames as potentially unsafe or to ignore embedded control characters.
- Capability inventory: The ingested filenames are passed to
adb pull,ls, andffmpegvia shell loops andawkscripts. - Sanitization: While basic shell quoting is used in some loops, the
awkscript that generates theffmpegconcatenation file (awk '{ printf "file %s\n", "\x27"$0"\x27" }') is vulnerable to filenames containing single quotes, which could lead to command failure or data corruption during audio processing.
Audit Metadata