rein: Skill for Claude Code

rein-plan

Use this after triage and before implementation when the work involves multiple files, multiple concerns, or a non-obvious execution order.

When To Use

• The task touches more than two files or modules
• The execution order matters (migrations before code, types before consumers)
• There are dependencies between subtasks
• The risk of a partial implementation is high
• A completed rein-interview spec bundle already exists and should drive the plan

When Not To Use

• The task is a single-file, single-concern change
• Triage already produced a clear, linear path
• The user explicitly asks to skip planning

Interview Handoff

Before planning, check whether there is a completed rein-interview artifact to consume.

Artifact priority:

1. If the user provides --from-interview <slug|path>, use that artifact.
2. If the user provides or references a rein interview handoff --to plan result, use its sourceResult and recommendedSkillInvocation.
3. Otherwise, look under .rein/specs/ for rein-interview-*/result.json.
4. If one clearly matches the current task, use it.
5. If several plausible artifacts exist, prefer the most recent completed one whose intent, desiredOutcome, or inScope matches the task.
6. If no relevant artifact exists, proceed with normal planning.

When a result.json artifact is available:

• treat it as the primary source of truth for intent, desired outcome, in-scope, out-of-scope, constraints, acceptance criteria, assumptions, technical context, and execution bridge
• use spec.md or the transcript only if result.json is insufficient
• do not silently drift away from the artifact without naming the conflict
• if the live repo facts conflict with the artifact, call out the mismatch in the plan
• if the artifact is missing core planning fields, stop and report that the interview bundle is incomplete rather than planning from a hollow result

Steps

1. Load the relevant rein-interview result.json first when available.
2. List every discrete subtask required to complete the work.
3. Identify dependencies between subtasks. Order them so each step has its prerequisites satisfied.
4. Write each step in change -> verify form.
5. For each step, state:
   • what changes
   • why this is the minimum sufficient change
   • what must stay untouched
   • what could go wrong
   • how to verify the step succeeded before moving on
6. Reject speculative abstraction, flexibility, or helper extraction unless the plan explicitly justifies it.
7. Identify the first safe checkpoint — the earliest point where the repo is in a valid state and work can be paused or reviewed.
8. Identify rollback boundaries — points where a partial revert is clean rather than destructive.
9. Flag any step that requires a decision the user has not yet made.