rivet-intake

Intake — ideas are external files, the tool only consumes input

Raw ideas live as files under .rivet/intake/ — versioned, re-routable, never lost. The tool's code and skills never change per idea; only these inputs do.

File shape (.rivet/intake/<slug>.md)

---
id: <slug>
source: raw | github#123 | gitlab#45 | jira:ABC-7
date: <YYYY-MM-DD>
---
<the idea, verbatim — do not editorialize the user's words>

## Refined (added during intake)
- Goal: <one sentence>
- In scope: …
- Out of scope: …
- Open questions: …

Process (RFC-2119)

1. Capture the user's words VERBATIM into the file first — messy is fine; never lose the original.
2. Run rivet route --file .rivet/intake/<slug>.md and report the mode + reason. MUST confirm with the user before proceeding (confirm-first).
3. If the idea is vague, ask clarifying questions BEFORE any code — one at a time, ≤5, each with a recommended answer. Append answers to the file's ## Refined section so the intake file stays the single record.
4. Then branch by mode:
   • research → investigate and report; write findings back under ## Refined. NO code changes.
   • quick → one delta + one test; create the task directly (rivet task create … --check …).
   • full-spec → hand off to the rivet-spec-author skill; the spec MUST link back to the intake id.
5. Tickets: when the source is Jira/GitHub/GitLab, copy the ticket body into the file and record the source id in frontmatter. Write-back to the ticket only if config intake.writeBack allows.