new-decision

name	new-decision
description	Scaffold a new decision file with the canonical structure — title, context, options, constraints, stakeholders, timeline, open questions — ready to feed to /decision:analyze. Use this when the user describes a decision verbally rather than handing over an existing decision file.

New Decision

Creates a structured decision document the framework analyses can operate on.

Procedure

1. Resolve output location. From $DATA_ROOT/config.json, read outputs.mode and outputs.local_root (or outputs.repo_path). The decision file goes to <root>/queue/<YYYY-MM-DD>-<slug>.md until it's processed (then analyze will move it into a workspace).

   If the queue directory doesn't exist, create it.

2. Interview the user. Walk through:

   • Title — short, decision-shaped ("Switch payments processor", "Take the contract role").
   • Context — what's the situation, why is this decision live now?
   • Options — list 2-5 concrete options. Reject "do something or don't" framings — push for at least 3 distinct paths.
   • Constraints — hard requirements (budget, timeline, ethical, contractual, capacity).
   • Stakeholders — who's affected, who has influence, who has veto.
   • Timeline — when does this need to be decided? When do consequences land?
   • Open questions — what's still unknown that the user wishes they knew.

   Ask one section at a time. Don't fire all six questions at once.

3. Write the decision file:

   # Decision: <Title>
   
   **Date drafted:** <DD/MM/YY>
   
   ## Context
   ...
   
   ## Options
   1. **<Option A name>** — <one-line>. <expanded>.
   2. **<Option B name>** — ...
   3. **<Option C name>** — ...
   
   ## Constraints
   - ...
   
   ## Stakeholders
   - ...
   
   ## Timeline
   - Decision deadline: ...
   - Consequences land: ...
   
   ## Open questions
   - ...
   

4. Confirm the file path and offer to immediately run /decision:analyze <path>.

Notes

• A decision file with fewer than 2 options can still be analyzed (e.g., "should I do X or not"), but flag this — many frameworks degrade when there's no comparator.
• Keep options balanced in detail. If one option has a paragraph and the others have a sentence, the analysis will be biased.