End-of-day engineering-learning harvest. Mines the day's Claude Code session transcripts for a repo and reports the TRANSFERABLE knowledge the user gained that day - portable principles, tools/commands/flags learned, traps to avoid, reusable domain facts - phrased so they help on a different project. Customizable per run via a focus lens (engineering, process/workflow, domain, product/judgment, or everything), passed as an arg or chosen interactively. Repo-scoped, reading every session that touched the repo that day, not just the current terminal. Output is printed in the terminal as numbered points with short explanations and writes no files. Trigger when the user types /eod or says "what did I learn today", "what did we learn", "what valuable knowledge did I gain", "end of day review", "eod", or asks for the engineering lessons / career-useful takeaways from a day's work in a repo.
Teach the user the recent work or a named topic from the ground up. A wise, incredibly effective tutor runs an incremental, mastery-gated session with a running checklist and quizzes, and does not stop until the user has demonstrated he understands the problem, the solution, and why it matters. Use when the user wants to be brought up to speed on what was built, a session's changes, a PR, or an unfamiliar area of the codebase. This is backward-looking (understand work that already exists). For forward-looking speculative planning, use grill-me instead.
Break a complex multi-step plan into the smallest atomic steps, render as a visual HTML artifact, then execute one step at a time with explicit user approval before AND after each step. Each atomic step is self-contained (its description, files, and verify check do not reference other steps) and has a single owner, a single verifiable outcome, and commit-sized scope. Decisions surface inline at the step that depends on them, never bundled upfront. Use when the user says "atomic plan", "/atomic-plan", "atomic-plan this", "break this into atomic steps", "smallest atomic pieces", "ask before each step", "ask question before each atomic step", "step-by-step with check-ins", or "let's do this slowly with check-ins". Different from /autoplan (parallel reviews) and /plan-eng-review (plan critique): atomic-plan executes an existing plan slowly with max oversight.