Concept Drill
Diagnoses your weakest sub-concept, drills it live via predict/break/confirm, leaves a tracker.
Getest · Werkt
Wat het doet
A 4-phase active-recall tutor: loads a prior progress tracker if you have one, diagnoses which small sub-concepts ('nodes') are wobbly with sharp predict-it questions instead of a quiz, drills only the weak nodes live in chat until they flip unprompted, then writes an updated tracker and a set of take-away FIRE exercises. Triggers on /drill [topic], /practice [topic], or phrases like 'drill me on X' / 'test where I'm weak on X'; explicitly does not trigger on passive 'teach me X' requests.
Testrapport
The trigger/anti-trigger boundary is unusually precise (explicitly refuses 'teach me X') and the drill loop (predict before reveal, drop-down-to-prerequisite on stall, honest wobbly/shaky/fluent grading) produced a genuinely diagnostic session in testing, not a generic quiz — the only wrinkle is it targets claude.ai's file/present_files environment, so Claude Code just writes the tracker/exercise files to the working directory instead of /mnt/user-data/outputs/.
Getest op: 2026-07-15 · Claude Code 2.x (agent harness)
Installatie
git clone https://github.com/Wamikmk/concept-drill.git cd concept-drill mkdir -p ~/.claude/skills cp -r skill ~/.claude/skills/concept-drill
Commando's en voorbeeldprompts
/concept-drillDiagnoses your weakest sub-concept, drills it live via predict/break/confirm, leaves a tracker.
Skills reageren op gewone verzoeken — geen commando's om te onthouden. Na installatie activeren prompts zoals deze de skill (in het Engels):
drill me on Bayes theorem until I actually get itfind where I'm weak on recursion and practice just thatquiz me with predict-first questions on binary search trees