The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses a vulnerability to indirect prompt injection as it converts untrusted audio input into text which is then analyzed by the model.
Ingestion points: Audio recordings provided via file paths or URLs as described in the Usage section of SKILL.md.
Boundary markers: Absent. The instructions do not define clear delimiters or provided warnings to the model to ignore potential instructions embedded within the transcribed content.
Capability inventory: The skill performs file system access, interacts with transcription services, and uses FFmpeg for audio processing.
Sanitization: Absent. There is no evidence of filtering or sanitization being applied to the transcript before it is passed to the summarization logic.
[EXTERNAL_DOWNLOADS]: The skill allows the agent to fetch audio content from arbitrary URLs, which entails downloading data from potentially untrusted external servers.
[COMMAND_EXECUTION]: The workflow specifies the use of FFmpeg for audio normalization and format conversion. This involves the execution of shell commands, which presents a potential risk of command injection if the input file paths or parameters provided by the user are not strictly sanitized.

meeting-transcription