deep-research

Performs Devil's Advocacy, Key Assumptions Check (CIA tradecraft), and bias audit on research findings. Produces actionable items for strengthening the analysis. Spawned by the research orchestrator after synthesis.

Extracts factual claims from source documents, builds the Analysis of Competing Hypotheses (ACH) matrix, triangulates evidence across sources, and flags uncorroborated or contradictory claims. Spawned by the research orchestrator after source rating and page caching.

Performs MECE decomposition of research questions, generates STORM stakeholder perspectives, formulates competing hypotheses, and produces multi-wave search queries. Spawned by the research orchestrator at the start of each research and between retrieval waves.

Autonomous single-agent deep researcher. Runs the full 11-phase research pipeline independently without user interaction. Spawns retrieval agents for parallel search and source evaluator for ratings, but handles scoping, triangulation, ACH analysis, critique, synthesis, and report generation itself. Use when delegating research to a background agent.

DEPRECATED — Orchestration now happens in the main session (see SKILL.md Step 5). This file is kept as a reference for the 11-phase pipeline architecture. Do NOT spawn this agent — the main session orchestrates directly via TeamCreate/SendMessage/Agent.

Conduct world-class research with 100-1000+ sources, competing hypothesis analysis, structured evidence management, and rigorous verification. Triggers: "deep research", "comprehensive analysis", "research report", "compare X vs Y", "analyze trends", "state of the art", "research on", "investigate". Do NOT use for simple lookups, debugging, or questions answerable with 1-2 searches.