# Agent slop

Agent slop is the low-value output that AI coding agents leave in a repository: duplicated code, redundant documents and commits too large to review. Nicholas Hopper measured it in a real repository and found that it is structural and historical, and invisible to a linter.

- **Published:** November 17, 2026
- **Updated:** November 17, 2026

## What Nicholas Hopper measured

Hopper analysed a monorepo that agents had built in 19 days across 740 commits. A linter found zero TODO, FIXME, HACK or XXX markers across 767 source files. By that measure the code was clean.

The repository's history said otherwise:

- 1,207,450 lines added and 42,157 removed.
- 101 near-duplicate QA documents in one directory, with names ending in "-final-reproof", "-2" and "-3".
- The most-edited file in the repository was a document, with 68 commits.
- Of 4,254 paths ever created, 32 were ever deleted.
- A median of 4 files per commit, a 99th percentile of 164, and one commit that touched 2,098 files.
- 351 of 767 source files had exactly one commit.

His conclusion: "Slop is structural and historical."

## What he built from it

Hopper built VDE, a desktop app that reads a repository's git history and reports fourteen signals, each oriented so that higher is worse. He refused to combine them into one score. The analysis uses no language model and no network, so the same commit always produces the same numbers.

VDE also generates a remediation brief addressed to a coding agent. One of its ground rules tells the agent not to create new files to describe the work.

## Related skills

- tree-sitter
- Electron
- [Claude Code](https://nicholashopper.com/skills/claude-code.md)

## Related projects

- [VDE](https://nicholashopper.com/projects/vde.md)

---

Canonical page: https://nicholashopper.com/notes/agent-slop  
Last updated: 2026-11-17  
Index of all pages: https://nicholashopper.com/llms.txt
