todo-agentlisted
Install: claude install-skill smorrel1/claude-code-skills
# To Do agent
Stephen's To Do note is 100KB long and he reads the first page. Two things follow from that:
work written to the bottom is invisible, and the note is the natural place for him to give
instructions, because it is already open.
## Inbound: he writes, agents act
Start any bullet in a To Do note with `claude` and then the instruction. The `@` is optional
(19 Sep 2026: dictation and the phone keyboard both swallow it):
```
@claude get me the costings from Annie and put them in the grant budget
Claude: get me the costings from Annie
claude get me the costings from Annie
```
Every five minutes the launchd job reads THE PINNED NOTE ONLY (Stephen, 21 Sep 2026:
`state/target-note.json`, currently "To Do 21st Sept 2026 Mac"), finds instructions it has
not seen, and hands
each one to the **Chief of staff** workspace with the section heading above it for context. The
Chief of Staff does it or delegates it, then records the result back on the line:
```
@claude get me the costings from Annie ✅ delegated to the grants workspace (16:20)
```
The trigger must **open** the bullet, after any list marker or number. Claude mid-sentence is
prose about the tool, not an instruction to it, which is what keeps the note's own "how to use
this" line from firing.
Dropping the `@` puts the trigger one word away from ordinary prose, so a bare `claude` has to be
corroborated: either punctuation straight after the word (`Claude: ...`, `Claude, ...`) or an
imperative verb next (`claude dra