The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted user data within database search queries.
Ingestion points: Untrusted data enters the agent context through the query_text parameter in multiple SQL scripts, including scripts/hybrid_search_fts.sql, scripts/hybrid_search_bm25.sql, and scripts/fuzzy_search.sql.
Boundary markers: Explicit boundary markers or instructions to ignore embedded commands are generally absent in the SQL function parameters, as they rely on standard SQL calling conventions.
Capability inventory: The functions perform database read operations (SELECT) and ranking across ALL scripts; no administrative, write, or file-system-level capabilities are exposed.
Sanitization: Several scripts include sanitization logic to mitigate injection; scripts/prefix_tsquery.sql uses regexp_replace and quote_literal to clean input tokens, and scripts/fuzzy_search.sql replaces LIKE wildcard characters (%, _) to prevent search pattern manipulation.
[EXTERNAL_DOWNLOADS]: The skill references documentation and benchmarks from trusted and well-known sources, as well as the vendor's own repository.
Mentions the author's benchmark repository on GitHub (github.com/laguagu/jev-rerank-bench) for performance evaluation.
References official documentation and tools from ParadeDB (paradedb.com and github.com/paradedb/agent-skills), which is a well-known service.
Links to the official pgvector repository's CHANGELOG for version and upgrade guidance.

postgres-semantic-search