Research Skill (Codex) — Step 2: Forked Agent Research

To validate feature requirements, you can fork parallel Codex agents to investigate alternative approaches, requirement validation, and risk analysis concurrently. This produces a consolidated codex research artifact for deeper reliability.

Generating requirement option sets involves running parallel agent roles to analyze feature research and output 2-3 distinct, user-selectable options. This eliminates rework by deriving choices directly from consolidated findings.

Forked agent research works by reading Step 1 local and domain research inputs, then running three focused Codex agent roles in parallel. Results merge into a consolidated codex research artifact and requirement option sets.

You cannot use this pipeline step if Step 1 outputs are missing, because the forked agents require reading prior local and domain research inputs to generate consolidated codex research and requirement options.

Using forked parallel agents provides deeper, reliable research by covering breadth and rigor across alternatives, validation, and risk simultaneously. A single agent cannot match this consolidated depth without rework.

Limitations include requiring existing Step 1 outputs in both local and domain research folders, and dependency on a multi-LLM cooperative pipeline. It cannot generate requirement options without these specific input paths.