What problem does it solve?
High-stakes research questions often get shallow, single-pass answers with unverified claims, circular sources, and hidden conflicts. This Skill enforces a contract-governed workflow that decomposes the question, gathers evidence in parallel, adversarially verifies every material claim, and only then synthesizes a cited answer.
Core Features & Use Cases
- Structured decomposition: Frames the decision, time boundary, and 3-6 non-overlapping sub-questions before any searching begins.
- Parallel investigation: Spawns one researcher work package per sub-question via subagents_orchestrate, each returning a claim ledger with source URL, date, authority, and primary-source status.
- Adversarial verification: A separate verifier agent maps every material claim to cited sources, requires two independent domains for high-confidence claims, detects conflicts, and returns VERDICT: PASS or FAIL with citation coverage metrics.
- Evidence-quality reporting: The final synthesis includes inline citations, explicit disagreement, and a footer reporting citation coverage, independent controlling domains, and conflict disposition.
- Use Case: Ask for a deep investigation into whether a market is worth entering; the workflow fans out investigators across market size, competitors, and regulation, then a verifier disproves weak claims before you receive a cited, conflict-aware synthesis.
Quick Start
Ask the agent to run a deep research workflow on your question, for example: do a deep research on the competitive landscape of vector databases with cited sources.
