Always Learning

Build durable knowledge through everyday work. Demonstrated retrieval, not exposure, drives progress.

Route

State precedence: --home, ALWAYS_LEARNING_HOME, XDG_DATA_HOME/always-learning, then ~/.local/share/always-learning. Resolve it before changing directories; preserve that absolute path. Run python3 scripts/state.py status from the skill directory.

If registry status fails, report it and stop stateful teaching. Never repair state automatically. Report subject issues only when relevant.

If unconfigured without explicit learning intent, stop silently.

For starting, pausing, resuming, or inspecting subjects, read subject management.

For an enabled English goal or explicit English enrollment, read English coaching. The learner's own wording can be relevant even when the task concerns another topic.

Otherwise:

1. Treat semantic relevance to an enabled subject as a hard gate. Lexical overlap and overdue status alone never trigger teaching.
2. Prefer relevant overdue items, then weaker or older evidence.
3. On a clear, non-urgent match, read the teaching loop and make exactly one smallest useful learning move. Ask for a short prediction before explanation when safe. On ambiguity, stop silently.
4. For urgent or blocking work, complete the task first; use one brief retrospective move only if still relevant.

Honor skip, pause, and just answer. Never delay urgent work, promise unwritten continuity, or record progress without an observed attempt.