shadcn/improve

Use your most capable model to audit your codebase and write plans for cheaper models to execute.

View on GitHub ↗Jump to charts ↓

Summary Information

Updated 8 minutes ago
Added to GitGenius on September 1st, 2026
Created on June 10th, 2026
Open Issues & Pull Requests: 16 (+0)
GitHub issues: Enabled
Number of forks: 406
Total Stargazers: 9,090 (+1)
Total Subscribers: 35 (+0)

Repository Insights (GitGenius)

Median issue/PR response: 16.0 hours
Mean response time: 2.6 days
90th percentile: 10.2 days
Tracked items: 5

Charts & Analytics

Fetching additional details & charts...

Issue Activity (beta)

Open issues: 4
New in 7 days: 0
Closed in 7 days: 0
Avg open age: 60 days
Stale 30+ days: 3
Stale 90+ days: 0

Recent activity

Opened in 7 days: 0
Closed in 7 days: 0
Comments in 7 days: 0
Events in 7 days: 0

Top labels

No label distribution available yet.

Most active issues this week

No issue events were indexed in the last 7 days.

Detailed Description

improve is an agent skill that audits codebases and writes implementation plans for other agents to execute.

The tool solves the problem of inefficient use of capable models by separating the intelligence-intensive work of understanding a codebase and deciding what matters from the execution work of implementing changes. It uses a capable model to map the repository, audit it across nine categories (correctness, security, performance, test coverage, tech debt, dependencies, developer experience, documentation, and feature direction), and write detailed plans. Those plans are then handed to cheaper models or humans for execution. The approach works by first performing reconnaissance to understand the repository's stack, conventions, build and test commands, and any existing design documentation. It then fans out parallel audits across the nine categories, with each finding carrying file-line evidence, impact, effort, and confidence scores. Before presenting findings, the tool re-verifies each cited location to filter false positives. The user selects which findings to plan, and the tool generates one markdown file per plan with exact steps, verification gates using the repository's own test and lint commands, and stop conditions for when reality diverges from expectations.

The tool suits teams that want to systematize code improvement without burning expensive model tokens on execution. It works well for repositories that maintain design documentation like architecture decision records, product requirements, or context files, since it ingests these to avoid re-flagging decided tradeoffs. The plans are plain markdown, so they can be executed by any agent or reviewed by humans before implementation. The skill integrates with any agent supporting the Agent Skills format.

The project maintains a workflow that supports iterative improvement: users can run a quick audit, select findings to plan, execute plans in isolated worktrees with diff review, and then reconcile the backlog in subsequent sessions to verify what landed and refresh what drifted. A branch-scoped mode lets teams audit only the changes in a pull request before submission.