The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it is designed to ingest and transform untrusted content provided by users or fetched from the web, which is then processed using powerful capabilities.
Ingestion points: The skill takes input from the user regarding technical topics and objectives in Phase 1. It also uses the WebFetch tool to retrieve external content when the user requests to 'turn this into a blog post' or similar.
Boundary markers: There are no instructions requiring the use of delimiters (such as XML tags or triple quotes) or specific 'ignore embedded instructions' warnings when handling user-provided or external data.
Capability inventory: The skill is configured with Write, Edit, and Agent tools in its allowed-tools metadata, allowing it to modify the filesystem and delegate tasks to other skills (like copywriting-hooks or copywriting-cta) based on potentially compromised input.
Sanitization: The instructions do not include any steps for sanitizing, validating, or filtering the ingested external content before it is used to generate article drafts or passed to sub-agents.

technical-article-writer