# Box Me In - are the limits around your next small project helping or squeezing?

> A playful check of the limits around one small project you are about to start. Type the project and 3 to 6 limits; mark each as a fact (the world sets it), a rule (someone else sets it), a habit (an old way of doing it that may not be a real limit) or chosen (a limit you set on purpose), and as fixed or bendable. The page counts them, draws them as slats in an SVG bench clamp whose jaw closes with the squeeze, and applies a fixed rule - free. One metered task, `box-limits`, reads each limit, turns it into a push question, says what room the limits leave and writes one move on the page's target. The page re-checks every reply.

## The rule (checked in order)
- paper-walls: habits are at least half of the limits. Move: test the first habit.
- vise-locked: 5 or more limits and at least 3 in 4 fixed. Move: loosen one fixed limit (a chosen one first, then a rule).
- wide-open: no limit chosen on purpose. Move: add a card from the page's deck (such as "use only what you already have").
- snug-fit: otherwise. Move: keep the first chosen limit.

## Contract
- Input (all strings): task "box-limits", project, limits (one line per limit: number, kind, fixed or bendable, words), tally (the page's per-limit lines, counts, verdict and move target), optional retry_note.
- Reply: one JSON object - task, verdict, headline, limits[{n, kind, reading, push}], space, move, closing.
- API tutorial in 8 languages: https://box-me-in.skillsafe.ai/api.html

## Free without signing in
The limit sorter, the clamp drawing (SVG and PNG), the page's verdict and move, four replayable samples (one per verdict), Markdown, CSV and JSON exports.

## Sources
- SkillSafe skill @lijigang/ljg-constraint - find the few constraints that really frame something, and sort them into world, rules and interpretations (hard, soft or self-set): https://skillsafe.ai/skill/@lijigang/ljg-constraint
- Inspired by the idea that a few well-chosen limits help creativity and learning.
- Model: gpt-terra, metered per run. Not professional advice.