resume

This skill implements the `/ar:resume` command for an autoresearch experiment workflow, letting a user pick up a previously started, paused, or context-limited experiment exactly where they left off. The problem it solves is continuity: long-running iterative experiments accumulate history in a branch and a results file, and re-establishing that context by hand is tedious and error-prone. The skill reads all of the prior state and produces a concise summary so work can continue without restarting.

The workflow is a small, read-oriented sequence. If no experiment is named, it lists available experiments via a setup script and shows each one's status (active, paused, or done) based on the age of a results.tsv file. Once an experiment is selected, it checks out the experiment's git branch, reads the config, strategy program, full results history, and the recent git log, then reports the current state: target file, optimization metric, counts of kept versus discarded versus crashed experiments, the best result relative to baseline, and observed patterns across change types. Finally it asks the user how to continue, offering a single iteration (/ar:run), an autonomous loop (/ar:loop), or a results review, and hands off accordingly.

The intended users are developers and researchers using this autoresearch agent framework to iteratively optimize code against a measurable metric such as latency. It is a thin orchestration/planning skill: its own logic is limited to reading history and routing to sibling commands, so its usefulness depends on the surrounding autoresearch tooling being present.