The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it analyzes Instagram comments. (1) Ingestion points: cluster_instagram_comments.mjs and rank_instagram_posts.mjs read caption and comment text. (2) Boundary markers: Absent in current processing scripts. (3) Capability inventory: Subprocess execution via yt_dlp, file writing to campaign directories, and network operations via cloud bridge. (4) Sanitization: Basic string normalization using cleanString and safeLower is performed before analysis.
[EXTERNAL_DOWNLOADS]: The skill communicates with Instagram and the vendor's APIs to retrieve research data. It includes a dedicated script for downloading video media from Instagram CDNs based on URLs identified in scraper manifests.
[COMMAND_EXECUTION]: Local scripts execute python3 -m yt_dlp for downloading reels. The skill implements a shellQuote utility to secure command-line arguments against injection. While generally safe, the use of scraper-provided identifiers in output file paths could theoretically present a path traversal risk if the upstream source data were maliciously manipulated.

instagram-audience-voice