retrieval-search-orchestration
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a strategy for handling multi-hop reasoning and retrieval over knowledge graphs, which involves processing untrusted user input and external corpus data, presenting a potential surface for indirect prompt injection. Ingestion points: The workflow processes user-provided queries (factoid, thematic, multi-hop) and external corpus documents used to build and query community summaries and entity neighborhoods (SKILL.md, methodology.md). Boundary markers: The methodology describes relevance gating using embedding similarity thresholds and community hierarchy levels to scope retrieval (methodology.md). Capability inventory: The orchestration includes generating structured graph queries (Cypher, SPARQL) and performing multi-step reasoning chains via LLM-as-controller or ReAct patterns (methodology.md). Sanitization: The design emphasizes provenance tracking, source annotation, and confidence scoring to validate retrieved information and identify conflicting evidence (provenance-patterns.md, rubric_retrieval.json).
- [NO_CODE]: The skill consists entirely of instructional markdown files and a JSON rubric, with no executable scripts or binaries included.
Audit Metadata