---
name: deslop
description: "Use right after writing new feature code or finishing a refactor in Smith: find LLM-generated Rust slop in the touched code and rewrite it to expert, type-driven, tested Rust without asking."
---

# Deslop

Turn plausible-looking generated Rust into expert-grade Rust.
Runs unattended: no questions, no approval gates, no stopping between passes.

## Scope

- Code touched by the current change (`jj diff --stat`, or the area the user names).
- Existing behavior stays; fix a bug only when proven and say so.
- Public API stays untouched.
- Second scope surfaces → list under Remaining, do not chase.

## Slop patterns

- Unverified APIs, hallucinated fns, wrong crate features.
- Overbroad abstractions: one-impl traits, generic soup, pass-through layers, dead code.
- Near-duplicate logic with renamed variables.
- Stringly typed domains where enums, newtypes, or validated structs belong.
- Excess `clone`, `Arc`, `Mutex`, boxing, dynamic dispatch, global state.
- Async faults: lock held across `.await`, blocking work in async paths, wrong bounds.
- `unwrap` | `expect` | `panic` | swallowed errors | vague errors where domain errors belong.
- Missing docs, missing tests, untested error paths, unverified security assumptions.
- Insecure input, path, shell, env handling.

## Rewrite bar

- Invalid states unrepresentable.
- Narrow APIs, explicit ownership.
- `Result` + `?`; document errors, panics, safety invariants.
- Concrete types until abstraction earns its keep.
- Delete unused code and needless layers.
- Behavior-boundary tests for success and failure paths.

## Work

1. Read touched code, call sites, tests, public surface.
2. Run scoped gates; keep pre-existing failures visible.
3. Rank findings by severity; fix P0 and P1 surgically, P2 when cheap, log P3.
4. Missing proof → add focused test before rewriting.
5. Run scoped gates again.

## Report

```markdown
## Status
complete | blocked | failed

## Findings
- `path:line` · pattern · evidence · P0 | P1 | P2 | P3

## Changes
- `path` · change · why

## Tests
- gates run, before and after

## Risks
- behavior, API, security, migration

## Remaining
- out-of-scope findings
```