Automatically audit and prune LLM-generated code artifacts, over-abstractions, and redundant tests across your codebase.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install review-llm-artifacts
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install review-llm-artifacts using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is LLM Artifacts Review?
As software teams increasingly leverage AI codegen, codebases often face a new kind of technical debt: LLM artifacts. These manifest as over-engineered patterns, redundant tests, verbose explanations, and dead code blocks. The review-llm-artifacts tool, a key asset within the Openclaw Skills repository, provides an automated framework to detect, log, and prune these AI-generated anomalies. By integrating this skill into developer environments, engineers maintain pristine code standards while benefiting from the speed of AI generation.
This utility scans changed files within your pull requests by default or sweeps the entire repository when required. It structures its assessment across four core categories—Tests, Dead Code, Abstraction, and Style—to construct a comprehensive, structured data overview. This makes it possible to audit code health reliably without introducing human review bottlenecks.
LLM Artifacts Review Use Cases
- Scanning codebases post-codegen to identify over-abstractions, excessive mock layers, or defensive overkill.
- Running automated PR gates against changed files to detect left-behind
TODOblocks or commented-out code. - Consolidating language-agnostic findings into a standardized JSON audit for continuous integration pipelines.
- Preparing files for downstream automated refactoring and cleanup agents.
How LLM Artifacts Review Works
- Determine Scope: Identifies whether to run in changed-files mode (using git merge-base to compare against main/master) or full scan mode (via the
--allflag). - Detect Languages: Parses file extensions (supporting Python, TypeScript, Go, Rust, Java, and more) to align the review with language-specific best practices.
- Execute Category Reviews: Launches parallel subagents (when supported) or runs sequential passes targeting Tests, Dead Code, Abstraction, and Style.
- Consolidate and Deduplicate: Integrates distinct sub-results into a contiguous ID matrix (
1..N) with zero gaps, enforcing an anti-confabulation policy to link every entry directly to a line of code. - Write JSON Report: Compiles metadata and detailed diagnostics into
.beagle/llm-artifacts-review.json. - Verify Integrity: Tests the JSON syntax and cross-checks the unique IDs, matching counts, and Git HEAD staleness before returning the final report summary.
LLM Artifacts Review Setup
Prerequisites
Make sure your git repository is initialized. The default scope requires a base branch (e.g., main or master).
Running the Skill
Execute the skill within your developer terminal to run an automated check on your workspace.
# Audit only files changed since merging with main
clippy-run review-llm-artifacts
# Audit the entire project workspace
clippy-run review-llm-artifacts --all
# Audit a specific directory path with forced parallel reviews
clippy-run review-llm-artifacts --parallel ./src/components
LLM Artifacts Review Data Schema & Taxonomy
The skill generates a comprehensive, parseable JSON report located at .beagle/llm-artifacts-review.json to organize its findings.
| Field Name | Type | Description |
|---|---|---|
version |
string | Version of the JSON schema (currently 1.0.0). |
scope |
string | The scan target scope, either all or changed. |
files_scanned |
number | Total number of source files analyzed during the sweep. |
findings |
array | Array of localized issues containing ID, file path, line number, category, description, and suggestions. |
summary |
object | Grouped counts categorized by severity, risk, fix safety, and module classifications. |
Example finding schema fragment:
{
"id": 1,
"category": "dead_code",
"type": "unused_import",
"file": "src/utils/helper.py",
"line": 42,
"description": "Unused helper functions left over from an LLM refactoring session",
"suggestion": "Remove the helper import",
"risk": "Low",
"fix_safety": "Safe",
"fix_action": "delete"
}
LLM Artifacts Review Advanced Features
- Parallel Category Processing: Leverages native multi-agent capabilities of Openclaw Skills to scan Tests, Style, Dead Code, and Abstraction in parallel, dramatically optimizing run times.
- Strict Verification Gateways: Integrates post-scan validation to guarantee JSON structure accuracy, check Git HEAD sync status, and verify deterministic count values.
- Downstream Fix Binding: The sequential
1..NID locking format interfaces cleanly with adjacent tools such asverify-llm-artifactsandfix-llm-artifactsfor one-click resolution workflows.
SKILL.md
Loading
Related Openclaw Skills
Remix v2 Forms Code Review
An automated code review agent skill designed to scan Remix v2 codebases for common form, mutation, and state management anti-patterns.
Remix v2 Meta, Sessions, Auth, and CSRF Code Review
An automated review skill for auditing Remix v2 codebases to prevent breaking meta tag changes, cookie security gaps, improper auth gates, and missing CSRF protections.
Xu Zhexun A-Share Investment Perspective
An advanced AI agent persona and quantitative-heuristic reasoning engine that replicates the practical A-share trading framework of private fund manager Xu Zhexun.
Openclaw Scheduler
A durable, SQLite-backed orchestration runtime for scheduling multi-step shell commands and AI agent workflows with automated retries and manual approvals.
respond-pr-feedback
An automated AI agent skill that posts contextual replies to GitHub Pull Request review comments and handles conversation thread resolution.
Remix v2 Data Flow Review
An advanced automated code review skill for analyzing and optimizing Remix v2 loader, action, and revalidation data flows.
Featured*
