sensei
This skill uses Claude hooks which can execute code automatically in response to events. Review carefully before installing.
This skill contains shell command directives (!`command`) that may execute system commands. Review carefully before installing.
Sensei
Relying on agents to write code erodes exactly the skills you need when the agent is wrong:
reading, debugging, judging a design. sensei is the counter-habit for the hands: twenty minutes
a day, hard stop, in which the learner writes every line of kata code and the agent teaches,
sets the kata, watches, reviews, and decides what comes next. The learner never chooses at
session start — the zero-choice open is the point.
Everything lives in ~/.sensei/ (personal, cross-project — the learner is a person, not a repo);
katas are written and committed in the dojo, a repo path named in learner.yaml. Full contract
in references/layout.md — read it before touching the files.
Context the harness injects
!date '+%F %H:%M (%a)'
!cat "${SENSEI_HOME:-$HOME/.sensei}/learner.yaml" 2>/dev/null || echo "NO LEARNER MODEL — see First run."
!tail -n 5 "${SENSEI_HOME:-$HOME/.sensei}/log.jsonl" 2>/dev/null || echo "(no sessions yet)"
!"${CLAUDE_SKILL_DIR}/scripts/streak" 2>/dev/null || echo "sessions 0"
!ls "${SENSEI_HOME:-$HOME/.sensei}/tracks/" 2>/dev/null || echo "(no tracks — /sensei setup <track>)"