| name | process-inbox |
| description | Read raw notes in inbox/, classify and label each one, draft what should be drafted in the author's voice, and open the results for review. Never publishes on its own.
|
Process inbox
Turn the backlog of raw thoughts in inbox/ into useful, voice-correct output — without ever
publishing on its own.
The flow
- Read each note in
inbox/.
- Classify it — give it a label:
blog — a post idea worth drafting
product-idea — a feature or product concept to capture and shape
task — something to do
memory — a fact worth remembering
discard — not worth keeping
- Act on the label:
blog → draft the post with the writing-style skill, run ai-voice-check, save to drafts/
product-idea → write a short one-pager (problem, who it's for, why now, smallest version) to ideas/
task → add it to your task list
memory → write a small note to memory/
discard → archive, no action
- Move the processed note to
inbox/archive/.
- Open the drafts for review. Nothing leaves without explicit approval.
The label is the routing
The label decides what happens to a note. If you run your inbox as files, the label is a line of
front-matter the skill adds. If you run your inbox as GitHub Issues, the same labels map to issue
labels — and a GitHub Action can route on them (see .github/workflows/process-inbox.yml).
Examples (see inbox/)
- A note about writing a blog on cloning yourself for AI → label
blog → drafted into drafts/.
- A note about a tool that makes your voice portable across AI apps → label
product-idea →
shaped into a one-pager in ideas/.
The rule
Drafting, classifying, capturing — always fine. Publishing, posting, or sending — never without the
owner approving that specific item.