What problem does it solve? Deep research on a new topic often stalls at shallow web searches, loses progress when a session ends, and lacks measurable criteria for knowing when you have reached expert-level understanding. This Skill runs an autonomous, self-regulated investigation loop that iterates on a research seed until PhD-level coverage criteria are met, checkpointing every cycle to disk so work survives context limits and session closures. ## Core Features & Use Cases - Self-regulated research loop: Iterates cycles of thread selection, source evaluation, and consolidation until five measurable closure criteria are met (coverage, Tier-1 sources, saturation, Munger inversion, cross-thread synthesis), with a configurable hard cycle cap. - Transparent source tiering: Classifies every consulted source into Tier 1/2/3 or auto-rejected, applying Weighted Decision Matrix (WDM) and Munger inversion to avoid bias and include dissenting voices. - Defensive checkpointing and resume: Writes per-cycle checkpoints, state, threads, findings, and gaps to memoria/investigaciones/<slug>/, with --resume and --list modes to continue across sessions. - Use Case: A researcher investigating episodic memory mechanisms in humans invokes the skill in mixed mode with a local papers folder, prioritizes authors like Tulving and Maguire, and receives a structured workspace with synthesis, action checklist, and documented knowledge gaps after the loop converges. ## Quick Start Ask the AI to run /recursive-research and answer its questions about your research topic, mode (web, local, or mixed), and preferred sources to begin the investigation.
