Inbox — agentic workflow ideas
The hottest surface in the scaffold. An idea lands here before it’s even a full sentence. Resist the urge to polish — that’s what zz-research/ is for.
Two capture shapes (both fine)
Section titled “Two capture shapes (both fine)”Running bullets — INBOX.md
: One-liners, trailing-question-mark proposals, links. No frontmatter. Append with a date prefix (- 2026-04-24: could we …?) and move on. Use this when typing a whole file would kill the thought.
Per-idea files — YYYY-MM-DD-topic.md
: Once a thought is a paragraph or more, give it its own file. Matches sibling naming under zz-research/, so the lifecycle is navigable chronologically.
Lifecycle
Section titled “Lifecycle” raw thought │ ▼ zz-inbox/ │ ├─► zz-research/ articulated enough to write about at length ├─► research/zz-challenges/ hard open problem worth multi-session thinking ├─► 02-stack/patterns/ crystallized recipe, ready for tier-2 voice ├─► /dev/null turned out to be noise — delete or archive └─► stays here still developing, or still worth looking at laterInbox entries should feel cheap to write AND cheap to delete.
What goes here vs. elsewhere
Section titled “What goes here vs. elsewhere”| Signal | Home |
|---|---|
| ”Could we …?” / “What if …?” about the scaffold | here |
| ”The workflow should …” suggestions | here |
| Bug report on a specific tool I use (Claude Code, Zellij, …) | zz-research/ — needs a full write-up |
| Hard problem I don’t know how to frame | research/zz-challenges/ |
| Feedback on how I collaborate with Claude | user-level memory (ask explicitly) |
Current entries
Section titled “Current entries”Entries surface in Starlight’s sidebar by date (newest first). Scan INBOX.md first for the unpolished pile.