garygentry/system1
Route closed judgements (pick one, rate this, yes/no) to a decision model instead of reading everything yourself. Typed, calibrated answers in ~300 ms for a fraction of a cent, via the decide CLI.
Load this skill before you read through a file, diff or directory of many items to judge each one yourself. Use it whenever the user asks you to classify, route, triage, label, filter, vet or give a verdict on a batch of items (tickets, reviews, commits, CI failures, log lines, the commands in a script, source files, rows), to pick the best match out of a long list of candidates (one package, file, helper or option out of hundreds), or to check a diff, log or test output against rules or acceptance criteria (is the task done, is this safe to commit), however small, since an independent verdict beats grading your own work, even if they never mention decisions or decide. It hands each item to a fast, calibrated decision model through the decide CLI (about 300 ms and $0.00003 per item), which is cheaper and more consistent than reading every item yourself. Not for generating text, counting, arithmetic, dates or exact matching.
Save a decision-model question set as a reusable spec in this repo, with thresholds, a default source and examples checked against recorded answers, or repair a saved spec that gives wrong or undecided answers. Use when the user wants to keep, save, name or reuse a decide question, turn a recurring judgement into a spec, write or edit a file in .system1/specs/, or fix a spec whose answers look wrong. For a one-off question, use the ask skill instead.
Set up or troubleshoot System 1, the decision-model toolkit. Checks the decide CLI, API key, repo egress consent and network access from this shell, and walks the user through fixing each problem with their consent. Use only when the user explicitly asks to set up, configure, check or troubleshoot decisions.