The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it fetches and processes episode metadata (titles, show notes, and transcripts) from an external API (https://app.aipodcast.ing/api/episodes).
Ingestion points: The scripts/ai_podcasting_client.py script retrieves episode objects from the remote API and presents them to the agent.
Boundary markers: The skill does not implement clear delimiters or instructions to prevent the agent from accidentally executing directives that might be embedded in the fetched content.
Capability inventory: The agent has access to local file reads and network uploads via scripts/aip_local_upload_helper.py, as well as the ability to modify or submit episodes via scripts/ai_podcasting_client.py.
Sanitization: Although the script performs basic HTML tag removal for previews, it does not sanitize data to prevent the interpretation of embedded instructions.
[DATA_EXFILTRATION]: The skill includes a utility, scripts/aip_local_upload_helper.py, designed to read local files and upload them to a remote service (app.aipodcast.ing) via presigned URLs. This is a primary feature for media management but represents a potential exfiltration vector if the agent is manipulated into uploading sensitive local files instead of intended media assets.
[EXTERNAL_DOWNLOADS]: The CLI tools perform network operations to communicate with the external endpoint https://app.aipodcast.ing for API transactions and file storage.

ai-podcasting