sink-research-orchestratorlisted

# Sink Research Orchestrator > **When to use**: Load this reference when sink subagents encounter unseen languages or new technique categories that require fresh research — not well-known sinks. > **Goal**: Produce the most complete, citable, and actionable analysis of a sink class or vulnerability family. Never stop at the first few hits. Read whole resources. Every finding must quote its source in structured JSON. --- ## Phase 0: Intent Clarification Confirm the target before research begins: | Question | Default | |----------|---------| | Language / ecosystem? | Whatever the target codebase uses | | Technique categories? | All relevant categories, including niche and newly observed ones | | Include well-known sinks? | No — prioritize unfamiliar, under-documented techniques | | Citation format? | Structured JSON per finding | | Output target? | A reference doc or sink catalog for the active language | | Depth level? | Navigate full posts / writeups, not summaries | If the request says “comprehensive”, “exhaustive”, or “most complete”, assume broad category coverage, JSON citations, and depth-first navigation. --- ## Phase 1: Parallel Research Swarm ### 1.1 Decompose the Topic Break the target into 8–12 orthogonal research lanes. Each lane should be independent. Example lanes: - Type / class confusion - Property / prototype injection - Race conditions and async escapes - Runtime / serialization / parser abuse - Filesystem / process edges - Framework-specific gadg