Unstale — Development skill for Claude Code
You don't know what you don't know.
How to install Unstale
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open vuca-dev/unstale and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Unstale does
You don't know what you don't know. Neither does your agent. unstale finds the breaking changes, fixes and new APIs in your dependencies that touch your code, for Claude Code, Codex and Cursor. Local, cited, free.
Alternatives in Development
- Sanitize — Detect and redact PII from text files — 15 categories (SSNs, credit cards, API keys, etc.), zero dependencies 3.1k ★
- Flow Parallel — Decompose and execute large changes, migrations, or multi-issue fixes in parallel with quality gates 2.8k ★
- Haft — Engineering decisions engine that know when they're stale 1.4k ★
README
unstale
**You don't know what you don't know. Neither does your agent.**
unstale keeps up with your dependencies for you. It finds the breaking changes, fixes and new APIs that touch your code, so your agent can use them. Most of those changes shipped after your agent's knowledge cutoff.
**Recent catches:** unstale flagged critical [Next.js security fixes](https://github.com/vercel/next.js/security/advisories/GHSA-2464-fhr4-w9xw) before the GitHub advisory database had them, and caught Supabase's Next.js auth blocks breaking on Next.js 16. Its fix for a deprecated AI SDK call in fumadocs was merged within a day.
unstale reads the official release notes, migration guides, docs and security advisories of the packages you use, distills each change once, and matches it against *your* code. Your coding agent (Claude Code, Codex or Cursor) then knows what applies here: new APIs worth adopting, breaking changes for your next major upgrade, deprecations, and the security and performance fixes in each version bump. Nothing else.
Why unstale
- Context-efficient by design. Changes are distilled centrally, once, for everyone. Your agent receives a small, ranked batch for your stack, not a pile of changelogs to read.
- Repo-aware. It reads your manifests, lockfiles and code, and ranks each change by where your code actually uses it.
- Source-cited. Every item links to the release note, migration guide, docs page or security advisory it came from. No blog takes, no guesswork.
- Zero noise is a valid answer. If nothing upstream matters to your code, your agents hear nothing.
- Private. Matching runs on your machine. Your code, paths and versions never leave it.
- No surprise edits. Your agent proposes, you decide. Decisions are remembered, so nothing is suggested twice.
- Alongside your update bot, not instead of it. Dependabot and Renovate tell you a new version exists. unstale tells your agent what that version, and the ones you have not ta
Related Skills
Mastering Java
I have a goal to do one leet code type problem every day, starting very simple and getting 1% harder each day
Controlloop
Know what your agents can do, before they do it. A local-first scanner and CI gate that reads agent artifacts,
Review Local
Review current working tree changes (diff) and suggest fixes
Claudetuner
Know when you'll hit your Claude, ChatGPT & Gemini usage limits — live 5h/weekly gauges, reset forecasts, and
Reduce Hallucination
A prompt that makes LLMs hallucinate less when they don't know.
Claude Find Skill
A Claude Code skill that finds skills you don't have yet
Related Agents
Part 13: Memory Bridge (Give Coding Agents Your Brain)
When you spawn Codex or Claude Code to build something, they start blind. They don't know your architecture de
Rig Debugger
Systematic root-cause debugger. Use when investigating a failing test, a production bug, a flaky behavior, or
Reference Scout
One-shot, read-only scout that finds how large, credible open-source projects solve a problem with the same te