The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill design creates a surface for Indirect Prompt Injection by ingesting untrusted content from social media platforms.
Ingestion points: Ingests data from X/Twitter (via fxtwitter API) and YouTube subtitles (via yt-dlp).
Boundary markers: Delimiters or specific safety instructions are absent from the provided skill templates.
Capability inventory: The skill possesses network access (WebFetch) and the ability to execute shell commands (yt-dlp, sed).
Sanitization: Basic text cleaning is performed using sed to remove formatting, but there is no semantic sanitization to prevent the LLM from following instructions embedded in the external content.
[DATA_EXFILTRATION] (LOW): The skill performs network operations to api.fxtwitter.com. This domain is not on the trusted whitelist, though it is a common community service for fetching social media metadata.
[COMMAND_EXECUTION] (SAFE): The instructions utilize standard system commands (yt-dlp, sed, grep, tr) to process media metadata and subtitles. These are appropriate for the skill's stated purpose.
[EXTERNAL_DOWNLOADS] (SAFE): The skill recommends installing yt-dlp, a well-known and reputable open-source tool, via standard package managers.

day5-fetch-and-digest