The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches podcast metadata from the iTunes Search API (itunes.apple.com) and downloads audio files from various external URLs provided in RSS feeds.
[COMMAND_EXECUTION]: The skill performs shell command execution via curl, mkdir, and uv run. It depends on local Python scripts (transcribe.py, remove_ads.py) and a uv project environment located at a specific hardcoded path: /Users/mtm/pdev/taylormonacelli/steamysquirrel.
[DATA_EXFILTRATION]: The skill exfiltrates data by uploading the processed MP3 files to litterbox.catbox.moe, an anonymous public file-sharing service. This involves network operations to a non-whitelisted domain.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content that may contain malicious instructions.
Ingestion points: External RSS feed XML (feed.xml) and transcribed audio text (transcript.json) generated from external MP3 files.
Boundary markers: No explicit delimiters or instructions to ignore instructions found within the transcribed content are used.
Capability inventory: The skill has the capability to execute shell commands, run local Python scripts, and perform network uploads.
Sanitization: The agent is instructed to 'read' the transcribed timestamped text to identify and cut ad segments; spoken instructions in the audio (e.g., 'Ignore previous instructions') could influence the agent's logic if they appear in the transcript.
[DYNAMIC_EXECUTION]: The skill uses python -c in Step 5 to execute a Python script string passed via the command line to filter the generated transcript.

podcast-ad-removal