The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a high surface area for indirect prompt injection attacks because it ingests and processes large amounts of untrusted content from the public web, including competitor landing pages, search results, and user reviews from third-party sites like Reddit, G2, and Capterra.
Ingestion points: External data is ingested via web search, web fetch, and specialized tools for reading ad libraries and social media posts as described in SKILL.md and rival-craft.md.
Boundary markers: The instructions do not consistently use strict delimiters or provide "ignore embedded instructions" warnings for all external content blocks, although the skill emphasizes fact-checking and source tracing to mitigate accidental obedience.
Capability inventory: The skill has broad execution capabilities, including importing data to the brand's knowledge base (gooseworks:knowledge_import), writing files (gooseworks:file_write), and configuring automated tasks (gooseworks:automation_upsert).
Sanitization: The skill implements proactive sanitization for data exports in linkedin-engagers.md by escaping formula-leading characters (e.g., =, +, -, @) in CSV files to prevent CSV injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill references and fetches data from several well-known technology services and platforms to accomplish its research tasks.
Evidence: Interacts with the Wayback Machine (web.archive.org), Hacker News (hn.algolia.com), and various official ad libraries (Meta, Google, LinkedIn, TikTok).
Context: These references target well-known services and are used for legitimate research functions; they are documented neutrally as they do not escalate the risk profile of the skill's own code.

research-competitors