The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests untrusted data from the web via the WebSearch tool and processes it into a newsletter.
Ingestion points: Web results from various AI news sources and research blogs are processed automatically.
Boundary markers: The skill uses markdown templates to structure output, but lacks explicit instructions to ignore embedded commands within the ingested news data.
Capability inventory: The agent can execute shell commands (npx tsx), write to the local file system, and perform network uploads to a CDN.
Sanitization: No explicit sanitization or validation of the fetched web content is described.
[COMMAND_EXECUTION]: The skill documentation references the execution of local TypeScript scripts (gather-ai-news.ts, process-research.ts, generate-newsletter.ts, optimize-images.ts) using npx tsx and system utilities like pngquant. These operations are used for content gathering, templating, and image processing.
[DATA_EXFILTRATION]: The optimize-images.ts script is described as having the capability to upload large images to a CDN and generate links. While this is a network operation that sends data externally, it is presented as a functional requirement for image hosting in newsletters.

ai-architect-newsletter