The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses a sensitive local file path (~/.secrets/aipodcasting/env) to retrieve authentication tokens. These credentials are transmitted to an external API (app.aipodcast.ing) which is not on the whitelisted domains list.
[DATA_EXFILTRATION]: The skill includes a local file upload utility (scripts/aip_local_upload_helper.py) that can resolve, read, and upload arbitrary local files to a remote server. This functional capability could be exploited to exfiltrate sensitive data if the agent is directed to process a malicious file path.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external API (such as episode titles, transcripts, and show notes) without providing boundary markers or sanitization, creating a surface for embedded instructions.
Ingestion points: scripts/ai_podcasting_client.py (via the run_list_episodes function) retrieves metadata and content from the app.aipodcast.ing API.
Boundary markers: Absent. The agent receives raw text content from the API responses.
Capability inventory: The skill can read local files, perform network uploads to remote presigned URLs, and modify remote episode data through PATCH and POST requests.
Sanitization: Absent. The code performs basic HTML tag stripping for display purposes but does not validate or sanitize content against malicious instructions.
[DYNAMIC_EXECUTION]: The included test suite (tests/test_ai_podcasting_client.py) uses the importlib module to dynamically load and execute the skill's Python scripts at runtime.

ai-podcasting