The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface. It ingests untrusted data from third-party web sources (product reviews, blogs, and news) and incorporates it into a generated HTML artifact. Without explicit instructions for the agent to sanitize or escape this content, malicious data from these sources could potentially include scripts that execute (XSS) when the user opens the resulting battlecard file. * Ingestion points: Web search results from competitor product pages, reviews (G2, Capterra), and news sites (SKILL.md Phase 3). * Boundary markers: Absent. * Capability inventory: File system write access to create the HTML artifact (SKILL.md Phase 5). * Sanitization: No sanitization or escaping protocols are defined in the execution flow.
[DATA_EXFILTRATION]: The skill encourages the aggregation of potentially sensitive corporate data (from CRM systems, internal documentation, and call transcripts) into a self-contained HTML file. While the file creation itself is a local operation, the skill's documentation suggests hosting the output on public web services (Netlify, Vercel), which represents a risk of accidental data exposure if the user does not implement proper access restrictions for the resulting URL.

competitive-intelligence