The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface during the podcast creation workflow where untrusted user documents are parsed and given to the agent to draft a conversational script.
Ingestion points: Data enters the agent's context from external documents (PDF, DOCX, Markdown, TXT) parsed via scripts/extract.py as detailed in SKILL.md.
Boundary markers: Absent. There are no explicit boundaries or instruction delimiters defined to prevent the agent from honoring instructions embedded within the extracted text.
Capability inventory: The skill executes local Python scripts (extract.py and google_tts.py) and interacts with the external Google TTS API via urllib.request.
Sanitization: scripts/extract.py includes a clean_text function that strips formatting markup (Markdown, HTML, code blocks, URLs), but it does not sanitize or validate the natural language content against malicious prompt injection payloads.
[COMMAND_EXECUTION]: scripts/google_tts.py invokes subprocess.run() to call ffmpeg for audio chunk concatenation.
Evidence: This execution passes arguments safely via a list structure without enabling shell=True, mitigating command injection risks and validating the static analyzer heuristic signal as benign.

google-tts