plan-feature

Plan Feature

Convert one shaped slice into the acceptance surface that implementation and review will share. The durable contract is the issue's behavior children, not narrative prose, parent strategy, or reviewer preference.

Inputs

• The shape packet or an equivalently bounded request.
• Repository instructions and relevant existing architecture.
• The issue identifier and title. Create a stable local identifier when none exists.

If the outcome, appetite, or smallest slice is still materially unsettled, stop and use $shape-strategy instead of planning competing interpretations.

Workflow

1. Inspect the active code paths and analogous features. Reuse facts; ask only for unresolved product decisions.

2. Freeze the ship contract as behavior children. One behavior is one observable, falsifiable claim with:

   • trigger or starting condition;
   • expected outcome;
   • concrete verification action and expected observation.

3. Treat every behavior as enforced. Set advisory: true only for an explicitly non-shipping diagnostic or follow-up.

4. Escalate one-way doors; decide the rest. Discovery settled the premise, but planning discovers choices the founder could not have anticipated. Record them in decisions, each marked reversible or one_way:

   "decisions": [
     { "decision": "Bake the SHA at build time rather than reading git at runtime.",
       "reversibility": "reversible" },
     { "decision": "Drop the legacy column in the users migration.",
       "reversibility": "one_way",
       "founder_answer": "Asked 2026-07-27: drop it, unread for two releases." }
   ]
   

   A one_way decision with no founder_answer blocks planning by name. Ask, then record what they said. Reversible decisions are yours to make — do not manufacture escalations for them, and omit decisions entirely when planning genuinely faced no such choice.

   One-way means expensive or impossible to unwind: schema migrations that drop or rewrite data, public API or URL shape, anything destructive, anything that commits an external contract. Trading away proof strength is also a product decision — demoting an enforced behavior's proof tier to fit an appetite is a choice about how much rigor to give up, not a technical detail.