The Agent Skills Directory

[COMMAND_EXECUTION]: The PostgresHybridSearch.hybrid_search method in resources/implementation-playbook.md contains a SQL injection vulnerability during query construction.
Evidence: The implementation iterates through filter_metadata and interpolates keys directly into the SQL string: where_clause += f" AND metadata->>'{key}' = ${len(params)}". If the keys in the metadata dictionary are derived from untrusted user input, an attacker could manipulate the SQL query structure.
[INDIRECT_PROMPT_INJECTION]: The HybridRAGPipeline and reranking templates in resources/implementation-playbook.md represent an attack surface for indirect prompt injection.
Ingestion points: _vector_search and _keyword_search methods in HybridRAGPipeline fetch external data.
Boundary markers: The templates do not demonstrate the use of delimiters or 'ignore' instructions for the retrieved content.
Capability inventory: The pipeline performs database operations and neural model inference (reranking).
Sanitization: No sanitization or validation of the retrieved document content (SearchResult.content) is performed before it is passed to the reranker or potentially a downstream LLM.

hybrid-search-implementation