ponder

What problem does it solve? Open questions often get answered from memory or a single search, leaving claims unsourced, rival explanations untested, and gaps unreported. This Skill enforces one investigative standard: decompose the question into retrievable sub-questions, source every load-bearing claim, test the strongest rival account, and disclose what stays unsettled. ## Core Features & Use Cases - Ledger-backed research sessions: A bundled CLI records leaves, sources, closes, rival sweeps, and checkpoints per round, rejecting batches with unresolved references and deriving the drafting scaffold from ledger state. - Source classification and hedging: Every source is classed as constitutive, attested, measured, or reported, and the check output flags which claims must be hedged based on evidence strength. - Structured answer output: Drafts carry [Sn] source markers, [~] markers for composed conclusions, and derived sections (Answer, Chain, Rival, Boundary, Open, Sources). - Use Case: Ask "Do multi-agent ensembles beat one strong model at matched compute?" and receive a researched answer where each claim traces to a retrieved source, the strongest contrary result is stated at its strongest, and unresolvable sub-questions are disclosed. ## Quick Start Ask the agent to research and answer an open question with sourced claims, for example: ponder whether tunneling a service through a third region adds significant latency over a direct path.