The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided audio recordings which are transcribed and then directly interpolated into a system prompt for structuring. This creates a surface where malicious instructions spoken in the recording could influence the agent's behavior or report output.
Ingestion points: The audio_path input in the voice_to_report function within SKILL.md leads to transcription that is used as prompt content.
Boundary markers: The prompt uses a simple text label ("Transcript:") but lacks robust delimiters or explicit instructions to ignore embedded commands within the transcript.
Capability inventory: The skill utilizes filesystem read access to open audio files and network access to call OpenAI APIs for transcription and completion, as defined in claw.json and SKILL.md.
Sanitization: There is no evidence of sanitization or validation performed on the transcribed text before it is processed by the LLM.
[EXTERNAL_DOWNLOADS]: The skill uses the whisper library, which fetches pre-trained machine learning models from OpenAI's official repositories via the whisper.load_model() call in SKILL.md. This targets a well-known service for legitimate functionality.
[COMMAND_EXECUTION]: The documentation includes standard setup instructions using pip install for required dependencies such as openai, whisper, and python-telegram-bot.

voice-to-report