Kabir Sharma

Three slash-command skills: /deep-research runs hierarchical parallel web research swarms (5 to 550 subagents), /boost-research adds adversarial fact-checking requiring 2+ source corroboration, and /turbo-boost races 3-5 parallel coders or investigators with blind testing and mutation breaking.

Users of Claude-style skill systems who need fast parallel research or verified coding output. The author is a student exploring GitHub, and the skills target anyone explicitly triggering /deep-research, /boost-research, or /turbo-boost on investigation or coding tasks.

Each uses a conductor architecture: the main process spawns ONE conductor subagent that autonomously runs the entire pipeline—spawning Flash subagents, enforcing a 2-minute hard timeout, tracking execution time, and returning a synthesized plain-English report without further main-process involvement.

It runs two phases: Phase 1 Parallel Recon floods the topic with 5-20 independent researchers, then Phase 2 deploys dedicated Fact-Checker and Devil's Advocate subagents that cross-examine and stress-test every claim. Only claims independently corroborated by 2+ sources survive synthesis.

No install dependencies, licensing terms, or costs are stated in the manifest. The skills are plain frontmatter-based definitions requiring a host environment that supports skill folders, slash-command triggers, and Flash-class subagent spawning.