/elegant-code-review

name: elegant-code-review
description: Review the current working diff against the elegant-code decision ladder and propose deletions, honoring the negligence floor.
usage: /elegant-code-review [--staged] [--base REF]

Elegant Code Review

Walk the current working diff against the `elegant-code` decision ladder. For each added hunk, name the rung that justifies it, flag any rung skipped, and propose concrete deletions. This is a per-change review. For whole-codebase cleanup, use `/unbloat`.

When To Use

• After writing or generating a change, before committing.
• When a diff feels larger than the problem it solves.
• When a new dependency appears in a lockfile or manifest.

When NOT To Use

• Whole-codebase dead-code sweeps: use `/unbloat` or `/bloat-scan`.
• Throwaway scripts where review cost is near zero.

Steps

1. Load the rubric: `Skill(conserve:elegant-code)`. 2. Collect the diff:

• Default: `git diff` (unstaged) plus `git diff --staged`.
• `--staged`: staged changes only.
• `--base REF`: `git diff REF...HEAD`.

3. For each added hunk, assign the lowest ladder rung that justifies it: 1. Need-to-exist 2. Builtin / stdlib 3. Native platform 4. Installed dependency 5. A few lines A new dependency is the last resort, below rung 5. If one was added, confirm it with `Skill(imbue:dependency-verification)` and check whether an installed dependency already covered the need. 4. Cross-check deletability with a real tool when available, treating output as candidates only:

   command -v knip >/dev/null && knip || echo "knip absent; skipping"
   command -v vulture >/dev/null && vulture . || echo "vulture absent; skipping"

5. Apply the negligence floor before proposing any deletion. Never propose removing input validation, authorization, data-loss handling, error handling, security paths, or accessibility. 6. Check completeness, not just excess. Flag a hunk that handles only the happy path: missing edge cases (empty, `None`, zero, boundary), missing negative or failure paths, or an obvious performance trap (quadratic loop, N+1 query). Minimal code that omits these is unfinished, not lean.

Output

A short table, one row per reviewed hunk:

| File:line | Chosen rung | Lower rung available? | Proposed action | |-----------|-------------|-----------------------|-----------------|

Close with a one-line verdict: `minimal and complete`, `trim suggested`, `new dependency unjustified`, or `incomplete: add coverage`. List proposed deletions, and any missing edge, negative, or performance coverage, as concrete diffs the user approves before anything changes.

Notes

• Read-only by default: propose, do not delete without approval.
• Composes `leyline:additive-bias-defense` for the burden-of-proof

verdict on each addition.