# Mirror Reps

> A practice coach for one close-up magic trick. Type the steps of the trick as you learned them, your level, the days and minutes a day you have, who will watch and where they will stand. The page sorts every step and estimates the practice it needs, for free; one metered run turns the steps into a chain of practice rungs drawn as padlocks, and after each run-through for real people a second run debriefs what happened and sends you back to the right rung. It never explains a secret method beyond the steps you typed, and never names a published trick, magician, book or brand.

URL: https://mirror-reps.skillsafe.ai/
API tutorial: https://mirror-reps.skillsafe.ai/api.html

## Free, in the browser
- Step sorting: each step is read as a set-up, a hidden move (double lift, palm, false shuffle, break, anything done "secretly"), open handling, a line ("say: ...") or a reveal. Hidden moves that can be seen from the side (palms, breaks, lifts, steals, loads, switches) are marked.
- The time gauge: your days x minutes a day against the page's practice estimate for the trick at your level (an estimate, not a measurement: per hidden move about 90 / 60 / 30 minutes for a first trick / a few tricks / performing often; 15 / 10 / 5 for each open move or reveal; less for lines and set-up; 15 or 30 extra per hidden move when watchers stand wide or at your sides; 3 minutes a step for full run-throughs). Fits, tight (under the estimate) or short (under 60% of it).
- The sightline fan: a top-down view of the table, you at the bottom, the watchers where you said, and the side wedges shaded by risk (low with everyone in front, medium in a wide half-circle, high with people at your sides or behind).
- The padlock chain: the page's own draft rungs (one per hidden move, the open moves and lines, a full run at the mirror, a first show for one person) until a ladder replaces them. Press a padlock when its rung passes; the next locked rung is marked "up next".
- The run-note cue spotter: counts words like "saw", "dropped", "knew how", "too fast", "laughed" in your notes so the debrief can be checked against them.
- Exports: Markdown, a practice-log CSV (rung, name, steps, minutes estimate, drill, pass check, passed), JSON, print; kept tricks (in the browser, or in your account when signed in).

## Metered (gpt-terra, two tasks)
- `ladder`: 3 to 8 rungs, each with its steps, a drill, an observable pass check and practice minutes (adding up to the smaller of your time and the estimate); one rung of its own for every hidden move; a full mirror run, then a first show for one person; one angle fix per hidden move that shows from the side; one line to say over each hidden move; a day-by-day plan covering every day once. Verdict by a fixed rule, checked in order: swap_trick (time under 60% of the estimate), tight (under the estimate, or people at your sides with hidden moves to guard), ready_in_time.
- `debrief`: findings quoted word for word from your run notes, each a caught, flash, fumble, pace, line, reaction or went-well; the rung to go back to; fixes; who to show next, what to focus on and how to check it; what to keep. Verdict by a fixed rule, checked in order: rebuild (two or more caught), drop_back (any caught, flash or fumble), perform_again.
- The page re-checks every reply: the verdict against the rule, every step in a rung, a rung of its own for each hidden move, the minutes total, one angle fix and one cover line per hidden move, the day plan, every quote against your notes, the rung to go back to against the rule, and no he/she, internal ids or stray numbers in the prose.
- Not coached: stunts that can injure (fire, sharp blades, spikes, piercing, swallowing, going under water, anything round the neck, weapons, chemicals). The page will not run either task on such steps, and the prompt answers swap_trick if called directly.

## Sources
- Built on the SkillSafe skill @jwynia/competency-builder (https://skillsafe.ai/skill/@jwynia/competency-builder): training that produces capability rather than knowledge, built as a ladder of observable competencies with feedback loops that surface gaps - narrowed to one sitting: practising one magic trick until it holds up in front of people.
- Why now: 31 October 2026 is the 100th anniversary of the death of the most famous escape artist of the last century; daily views of his English Wikipedia article rose from about 2,400 in mid-September to over 10,000 on 20 October 2026 (Wikimedia pageviews API).