Generate a terse research-style daily report for the current project. Runs almost entirely in a background agent — the foreground only resolves paths and the Notion target, then a background worker reads today's session transcripts + git activity, drafts the report to <project>/.claude/daily-reports/YYYY-MM-DD.md, and (if configured) syncs it as a dated child page placed at the TOP of a Notion parent page. Use when the user says "今日の日報書いて", "daily report", "今日のまとめ", or asks to summarize today's work in this project.
Run non-trivial implementation work as a two-phase agent team — investigator-planner produces PLAN.md and task decomposition, then per-task implementers + reviewer + smoke-tester spawn in parallel under auto mode and coordinate via SendMessage. Use whenever the user asks to "implement X", "build X", "add feature Y", "refactor X", "migrate X to Y", "リファクタして", "移行して", "agent team で進めて", "マルチエージェントで実装", "並列で実装", or "実装進めて". Size floor — only use when the task touches ≥3 files OR has ≥2 distinct phases (plan + execute). For one- or two-file edits, just do it inline.
Set up event-driven pueue monitoring (Monitor tool + jq state poller) so background pueue jobs notify Claude only on state transitions. Use when the user runs long pueue chains (training, conversion, copy) and wants Claude to react only on failure/completion without burning tokens on idle polling. Also covers the two pueue+Claude foot-guns we hit (argv tokenization when wrapping commands in `bash -c` / `bash -lc`, and the `Done.result` JSON shape).
Applies default matplotlib style for all figures: rcParams, color palette (single-accent monochrome), and conventions for bar charts, legends, annotations, and cross-figure consistency. Use whenever creating any matplotlib plot or visualization.
After a rosbag2 collection session in dxl-teleop-bilateral, copy the recorded task directory from local_dataset/ into the canonical rosbags/ tree, run the rosbag2->LeRobot conversion via temporal_force_encoding's pixi task, and validate the resulting dataset (action present, videos present, frame counts consistent). Use this whenever the user has just finished collecting episodes and wants the LeRobot dataset built and sanity-checked.
Methodology for iteratively improving agent-facing instructions (skills / slash commands / CLAUDE.md / code-gen prompts) by having a bias-free executor run them and evaluating two-sidedly (executor self-report + instruction-side metrics) until improvements plateau. Use after creating or revising a prompt or skill.
Fast, reproducible scientific Python environments with pixi - conda and PyPI unified
Uses typer for all Python CLI argument parsing instead of argparse. Use whenever creating a Python script that accepts command-line arguments, options, or subcommands.