LLM knowledge cutoff compensator for AI coding agents. Corrects outdated versions, deprecated APIs, renamed packages, and removed CLI commands — so your agent stops confidently using last year's code.
| Knowledge Corrections Renamed packages, changed APIs, new config formats, current model IDs |
Detect Hook Scans prompts and injects only relevant corrections automatically |
Multi-Platform Claude Code, Codex CLI, Gemini CLI from a single install |
| Version Verification Never caches versions — always verified via package manager |
Auto Update Checks registry for new corrections on each run |
Zero Config One command to install, corrections work immediately |
npx knowpatch
Select platforms, done. Corrections start injecting into your agent automatically.
User prompt → Detect hook scans keywords → Matches correction file tags → Injects relevant entries into agent context
The hook reads each prompt and matches against correction file tags. Only relevant entries are injected — not the entire correction set.
| File | Covers |
|---|---|
frontier-models.md |
Claude, GPT, Gemini model IDs |
open-source-models.md |
Llama, Mistral, DeepSeek, Qwen, GLM, Kimi |
cli-tools.md |
shadcn, tailwind, eslint, vite |
frameworks.md |
Next.js, Svelte, Nuxt, Django, FastAPI |
javascript.md |
Zod, React, TypeScript |
python.md |
Pydantic, Ruff, uv, pip |
platforms.md |
Supabase, Vercel, Cloudflare |
runtimes.md |
Node.js, Bun, Deno, Python, macOS |
Running knowpatch with no arguments opens an interactive menu. Subcommands are also available:
knowpatch install # install skill + hook
knowpatch update # sync installation, check for updates
knowpatch uninstall # remove skill and hooksOptions
| Flag | Description |
|---|---|
--scope user|project |
Installation scope (~/ or ./) |
--platforms claude,codex,gemini |
Target specific platforms |
--force |
Overwrite existing (install only) |
--reconfigure |
Re-select platforms (update only) |
bun install && bun run dev # dev mode
bun run build # build
bun test # testMIT