version: "1.0.0" name: refactor-small description: >- Identifies a single readability or maintainability problem in the Exis codebase and refactors it for better single responsibility, DRY, KISS, YAGNI, and a sound balance between concise and descriptive. Use when the user invokes /refactor-small or asks for one small, scoped readability/maintainability cleanup. disable-model-invocation: true

One scoped readability refactor

Find one real readability/maintainability problem and fix only that. This is a surgical cleanup, not a rewrite. Behavior must stay identical.

Follow the principles in AGENTS.md ("Readability and maintainability") — they are the source of truth and override anything here on conflict.

Scope rules

• Exactly one problem, one concept. If you spot several, fix the highest-impact one and (optionally) note the rest for later. Do not batch.
• No behavior change. Pure refactor: same inputs → same outputs, same side effects, same public API unless the user explicitly asked otherwise.
• No performance regression. Refactors must not sacrifice performance — avoid extra allocations, re-renders, or work on hot paths in the name of readability.
• Small blast radius. Prefer changes contained to a function, component, or a tight cluster of call sites. If it touches many files or needs design decisions, stop and switch to Plan mode instead.
• Match existing patterns before introducing new ones. Consistency with nearby code beats personal preference.
• No opportunistic churn — no reformatting, renaming, or reordering unrelated code.

Workflow

1. Pick a target. Use the candidate-finding tactics below, or honor the user's pointer (file, function, "this"). Confirm the target is genuinely worth touching — skip if the code is already clear.
2. Name the problem in one sentence and which principle it violates (SRP / DRY / KISS / YAGNI / naming / clarity).
3. Refactor minimally. Apply the smallest change that resolves it. Read the file(s) first; preserve types, invariants, and control flow.
4. Verify (see commands). Run the checks relevant to what you touched.
5. Report using the output template — what, why, and proof it's safe.

What to look for

Avoid the reverse failure too: don't compress code into something terse-but-cryptic. Aim for the balance in AGENTS.md — descriptive enough to avoid guesswork, concise enough to read fast.

Finding candidates (when the user didn't point at one)

• Grep for smell signals: // TODO, repeated literals, long match/if chains, any, unwrap() clusters, duplicated css({...}) blocks.
• SemanticSearch for "the function that does X" when prose-described.
• Prefer authoritative/hot files so the cleanup matters.

Output template

**Refactor:** `path:line` — [problem in one sentence]
**Principle:** [SRP | DRY | KISS | YAGNI | naming | clarity]
**Change:** [what you did, 1–2 lines]
**Safe because:** [why behavior is unchanged + which checks passed]

Show the key before/after with a code citation when it aids review.

Do not

• Touch more than one concern, or fan out across many files.
• Change behavior, public APIs, or test expectations to make a refactor "work".
• Reformat or rename unrelated code.
• Add abstractions for hypothetical future needs.
• Trade performance for readability (extra allocations, re-renders, or hot-path work).
• Proceed when the fix needs design decisions — switch to Plan mode and discuss first.