The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted audio data and passes the resulting transcription to external LLM tools (Claude or GitHub Copilot) for summarization and meeting minutes generation.
Ingestion points: The script scripts/transcribe.py ingests transcription segments derived from external audio files or URLs.
Boundary markers: The skill uses simple text separators (---) in the prompt construction but lacks robust instructions to the LLM to ignore or treat transcription content as untrusted data.
Capability inventory: The skill possesses capabilities to write files to the local file system and execute local CLI tools (claude, gh).
Sanitization: No filtering or sanitization is performed on the transcribed text before it is interpolated into the LLM prompts in scripts/transcribe.py.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run to execute local binaries including ffmpeg, ffprobe, python3, claude, and gh for audio processing and AI integration.
[DYNAMIC_EXECUTION]: The scripts/transcribe.py script attempts to dynamically install missing Python dependencies (rich, tqdm) using pip install at runtime if they are not already present in the environment, modifying the execution environment dynamically.

audio-transcriber