# Next

> Coach the user on what to fix first — rank the audit findings into a single, ordered to-do list by score impact and effort, and explain why each item is where it is. Read-only; advises, does not change anything. Use when the user asks what to do next, what to fix first, what matters most, how to prioritize, or where to start after an audit.

- Skill: `hainrixz/next` (Agent Skill)
- Install (CLI): `npx skillmds@latest add hainrixz/next`
- Raw SKILL.md: https://api.skillmd.com/api/skills/hainrixz/next/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: Hainrixz (https://skillmd.com/u/hainrixz)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/hainrixz/next

---


# /claude-db:next

The coach. Turns a pile of findings into one ranked, do-this-next list with a reason for each item. Read-only — it prioritizes, it never writes or migrates.

`$ARGUMENTS` = `[findings.json] [flags]`. Use the most recent audit's findings, or a findings JSON path. If no audit has run, say so and suggest `/claude-db:audit <target>`.

## Ranking (in order)
1. **Capped severity-5 first.** Any `status:"fail"` + `severity:5` on a relied-on axis caps that score at band F. These come first regardless of anything else — fixing them is the only way to lift the cap. (`needs_api` and `confidence:"speculative"` never cap, so they never jump the queue this way.)
2. Then by **severity × axis-weight** — the finding's severity times the weight its module carries in the relevant category/axis (from `references/scoring-model.md` / the active `PROFILES` profile in `score.mjs`). A high-severity finding in a heavily-weighted category (e.g. Indexación at w20, Integridad referencial at w16) outranks the same severity in a light one (Naming at w6).
3. Then by **fixability**: `auto` > `proposed` > `advisory` — surface the changes the tool can safely make for you before the ones needing judgment or that it will never write.
4. Then by **least reproduction cost** — break ties toward findings whose `verification.reproduce` is cheapest to confirm (Tier-0 static check before a Tier-1 query before anything `needs_api`).

## Output
A numbered list. For each item: the finding (id + one-line title), which score/axis it moves and roughly how much (banded magnitude high|medium|low — never a fabricated %), its fixability class, and the next concrete action ("run `/claude-db:fix --category keys`" / "run `/claude-db:migrate <file>`" / "needs a live DB: set `$DATABASE_URL` and re-audit at Tier 1"). Honor `--axis` and `--top N`. Respond in the user's language (EN/ES).