AI-SLOP Detector skill
A check for AI-generated code: empty functions, imports of things that do not exist, and docs that oversell.
- 113 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The tool reads project code and sends nothing out; with `--read-only` it writes no files.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/ai-slop-detectorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The tool reads project code and sends nothing out; with `--read-only` it writes no files.
This is third-party code. Review the repository files before installing.
What it does
AI-SLOP Detector finds what an ordinary linter misses: functions with a plausible but empty body, imports of modules that do not exist, copy-pasted logic, unreachable code and comments that promise more than the code does. It runs fully offline with no keys or model downloads, and the same input produces byte-identical output. Commands include `scan`, `review`, `pulse` and `sweep`, with JSON output, a `.slopconfig.yaml` configuration and a `pre-commit` hook. The repository ships a Claude Code skill that teaches the agent to run a structured check after each edit, plus MCP and REST interfaces. If nothing was measured, the report says so instead of returning a clean verdict.
Who it is for. Teams that accept code from AI agents and want to filter stubs and invented dependencies automatically before `review`.
Good fit when
- An agent reported a finished function whose body is `pass` and a comment.
- Before merging, only the changed code is checked through `review`.
- A repository health score is needed to prioritize cleanup.
- The check goes into `pre-commit` or CI with a `--fail-threshold` limit.
Not a fit when
- You need a security or vulnerability `scan`: the tool looks for structural defects.
- The code is in languages it does not parse and needs precise type analysis.
- The project calls for subjective criteria such as style.
Example request
Check the branch changes through `review` in JSON and list stub functions in descending severity.Limitations
The tool catches structural signs of stubs and exaggeration but does not prove the logic correct, so a clean report does not replace tests and `review`. Rules are heuristic and false positives are possible on unusual patterns; thresholds are set in `.slopconfig.yaml`. The main analysis targets Python and other languages get narrower coverage. The Claude Code skill needs the `ai-slop-detector` package installed separately with pip. The 3.x versions change quickly, so pin the JSON format to a version number.
How to disable. Delete the skill from the agent's skills directory and run `pip uninstall ai-slop-detector`. Remove the `pre-commit` hook from the configuration.
FAQ
Are API keys needed?
No. The core runs offline, downloads no models and sends nothing out.
How do `scan` and `review` differ?
`scan` analyzes the whole project as a baseline, `review` looks at changed code. The README names `review`, `pulse` and `sweep` as the main agent surfaces.
How do I add it to `pre-commit`?
The repository ships `.pre-commit-hooks.yaml`; add the hook to your project's `pre-commit` configuration.
What does NOT EVALUATED mean?
The report states nothing was measured, for example all files failed to parse. Such a run does not pass the `--fail-threshold` limit.
Is there MCP?
Yes, one `scan` service backs the CLI, MCP and REST. The REST server listens on `127.0.0.1` only.
Discussion
No questions or reviews yet.
Related
SkillsA CLI that finds traces coding agents leave in code, like filler comments, swallowed errors and fake imports, and scores it 0–100
Agent Reliability Analyzer statically scans AI agent code, finds reliability gaps with 340 plus rules and applies fixes
Two skills for Codex and Claude Code: a read-only repository audit and an experiment loop with saved baselines.
Validates Agent Skills packages: structure, links, token weight and instruction quality, with optional LLM-as-judge scoring
A static linter for agent instructions: finds ambiguous tool descriptions and schema gaps before a run, with no model or network
A skill that reads a repository once and installs checks: code rules become gates that fail a commit or a CI run.
README in short
Describes the check principles (determinism, honest reporting of nothing measured, tested tests), versions 3.10 to 3.13 and running from the CLI.