The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines the SearchableEvent interface and its implementations, which ingest and index human-authored text from various Nostr event kinds (e.g., TextNote, Metadata, LongTextNote) for search purposes.
Ingestion points: Raw data from event fields such as content, subject, summary, and description enter the agent's context through the indexableContent() method defined across 126 classes in references/searchable-kinds.md.
Boundary markers: The skill specifies that human-meaningful fields are joined with newlines (\n) or spaces ( ) without utilizing specific delimiters or "ignore instructions" warnings to separate user-generated content from indexing logic.
Capability inventory: The skill describes instructions for SQLite FTS5 virtual tables and filesystem-based tokenizers, as well as parsing NIP-50 search queries.
Sanitization: The documentation explicitly notes that "no markup stripping is performed," allowing raw markdown and other structured text to be indexed without sanitization, creating a surface for indirect prompt injection if the search results are later processed by an LLM as instructions.

searchable-events