/note

Frictionless note-to-self: you dump a thought, the skill files it. You never have to pick a category — but you can override one when you want.

Categories (fixed, small on purpose)

• skill — "this could be a Claude Code skill" candidates
• mcp — "this could be an MCP / tool server" build ideas
• todo — things to do or fix
• note — everything else (observations, decisions, reminders)

One file per category: ~/.claude/notes/<category>.md (create with a # <category> header if it doesn't exist).

Decide what the invocation means

1. No argument → LIST all: for each ~/.claude/notes/*.md, show its entries under the category name. If none exist, say the notes are empty.
2. Argument is exactly a category name (note / skill / mcp / todo) → LIST just that bucket (say it's empty if the file is missing).
3. Argument begins with <category>: (explicit override, e.g. skill: ...) → CAPTURE into that category, verbatim (text after the colon).
4. Otherwise → CAPTURE with auto-classification (below).

Capture

• If no category was given, classify the note into exactly one of {skill, mcp, todo, note}:
  • building/making a skill or slash command → skill
  • building an MCP / tool server → mcp
  • an actionable task (fix, bump, email, check, send…) → todo
  • otherwise → note

• Append exactly one line to ~/.claude/notes/<category>.md (create it with a # <category> header first if missing):

  - <today's date YYYY-MM-DD> — <the note, verbatim>

  (Use date +%F if unsure of the date.)

• Report where you filed it — e.g. "Filed under skill: …" — so a wrong guess is obvious and easy to re-file.

Keep it simple

Append and list only. Don't reformat or "improve" existing entries; one line per note. If the category is ambiguous, pick the closest and say so — the user can re-file by editing the file.