The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided URLs using WebFetch and capture_website to extract facts, quotes, and visual cues for podcast scripts.
Ingestion points: Website content fetched via WebFetch and visual data from capture_website are introduced into the agent context in SKILL.md.
Boundary markers: The instructions lack explicit requirements for using delimiters or "ignore embedded instructions" warnings when processing the fetched content into dialogue.
Capability inventory: The skill possesses powerful downstream capabilities including media generation (generate_reference_video, generate_speech), media analysis (analyze_media), and file uploading (upload_asset).
Sanitization: No sanitization, escaping, or validation steps are defined for the external content before it is interpolated into the script generation prompts.
[DATA_EXFILTRATION]: The skill implements a workflow to circumvent platform limitations by using the upload_asset tool to convert local files provided by the user into public URLs.
Evidence: In the 'Local images on Claude Desktop' section, the skill instructs the agent to ask users to attach image files which are then processed via upload_asset to obtain a public_url for use in remote generation tools. This design creates a data flow from the user's local environment to public cloud storage, representing a potential exposure vector for sensitive media.

podcast