/idea. Quick capture into project inbox

Drop ideas, half-formed thoughts, junk, or maybe-junk into a project-scoped inbox without context-switching the current task.

Tone & Voice

Zero-friction primitive. NEVER ask clarifying questions on a valid /idea call. that defeats the entire point. The whole reason this skill exists is to short-circuit context-switching. If the body is empty, ONE single-line prompt ("what's the idea?") is OK; otherwise capture verbatim and acknowledge with one line max.

Conversation context (prior)

Prior conversation: N/A. /idea is a one-shot command. The body after /idea IS the entire input. Don't pull context from prior turns to interpret or reframe the idea. capture it as-is.

Routing rule

1. Read memory/state/active_project.txt.
2. If non-empty AND memory/projects/<active>.md exists → route there.
3. Else → route to memory/projects/side-quests.md (default incubator per spec).

Procedure

1. Parse text: everything after /idea is the idea body. Trim whitespace. If empty, ask: "what's the idea?" (one line, don't ramble).
2. Determine target file via routing rule above.
3. Read target file, locate ## Inbox heading section.
4. Append entry in this format:

   - {YYYY-MM-DD HH:MM}. {idea text}

   • Use your local timezone for the timestamp.
   • If section currently has the placeholder (empty. awaiting...) text → REPLACE it with the new entry as the first list item.
   • Otherwise → append as a new bullet at the bottom of the Inbox section (before the next ## heading).
5. Update last_updated in target file's frontmatter to today's date.
6. Confirm: one short line, format:
   • 📂 idea → {project-name}: "{first 50 chars of idea}{... if longer}"
   • If empty active project → 📂 idea → side-quests (no active project): "..."
7. Skip semantic reindex: inbox entries get picked up on the next nightly reindex. Don't burn time for a one-line capture.

Edge cases

• Multi-line idea (idea text contains newlines): collapse to single line with ; separator. Chat messages mostly come single-line anyway, but the heuristic handles paste-edge-cases.
• Active project file missing: log warning, fall back to side-quests.md.
• Inbox section missing in target file: warn ("project file structure broken. falling back to side-quests"), append to side-quests instead. Don't try to repair the broken file mid-flow.
• Idea references a specific project via @-mention (e.g. /idea @project-alpha someconcept): override active-project routing, target the @-mentioned project. Validate the project exists; else error.

Graduation hook

When an idea entry triggers a longer thread (the operator explicitly references it later, or it accumulates >3 sub-bullets via subsequent /idea or in-context expansion), surface the question: "this looks substantive. graduate to its own project file? (memory/projects/<slug>.md)". See the project skill (not bundled in this skeleton) § Graduation criteria for the 4-condition test.

What this is NOT

• Not a TODO list. there's a task tracker for that. Ideas are unscoped musings; tasks are scoped commitments.
• Not a journal. that's the natural session log. /idea is for "noise that might be signal" only.
• Not a memory-write. doesn't touch Preferences/Decisions/Learnings. Pure inbox capture.

Example flow

> /idea what if we wrap the daily digest cycle output as a chat notification
📂 idea → project-alpha: "what if we wrap the daily digest cycle output as a..."

> /project clear
> /idea note-taking template for the weekly review
📂 idea → side-quests (no active project): "note-taking template for the weekly..."

> /idea @project-beta reduce scope on phase 3-4
📂 idea → project-beta: "reduce scope on phase 3-4"

Pairs with

Pairs with the project skill (not bundled in this skeleton) (the project context switcher), same filesystem-projects layer.