The trusted registry for AI instructions
A small, sealed catalog — not a dump of every prompt on the internet. 42 files we'd actually install, each with a public scan report.
Start with a pack
All collections →Coherent sets that work installed together — the fastest way to try Waxmark without wading through a firehose.
Agent starter pack
One of each format, working together: commit discipline (rule), project instructions (AGENTS.md), a rigorous review prompt, and a changelog skill. Works with any language and any tool.
React & TypeScript pack
A full front-end stack of rules that don't fight each other: React patterns, TypeScript standards, Next.js, Zustand state, React Query data fetching, and Jest testing.
Python services pack
Modern Python back-end rules: language best practices, FastAPI, Pydantic models, SQLAlchemy, pytest, and Docker packaging.
Data science pack
Analysis-ready rules for the scientific Python stack: pandas, NumPy, scikit-learn, and Matplotlib.
What the seal means
How scanning works →1 · Scanned on arrival
Four stages inspect every file: invisible Unicode, prompt-injection phrasing, exfiltration sinks, and embedded secrets.
2 · Sealed or quarantined
Failures never go public — a human reviews them. Warnings publish with the full report visible. Every badge links to what we found.
3 · Re-scanned on every edit
New versions re-scan mandatorily, so a file can't turn malicious after you trusted it.
How to use a sealed file
Once a file has the seal, it is ready to copy into your repo. Three steps:
Step 1
Pick a pack or browse
Start with a coherent set, or open one sealed file from the catalog. Every card links to the scan report.
Step 2
Copy or export
Native formats for Cursor rules, AGENTS.md / CLAUDE.md, skills, and plain prompts.
Step 3
Install and re-check
Drop it into your repo. Edits on Waxmark re-scan automatically, so a trusted file cannot quietly turn hostile.
Rules
Always-on constraints for coding agents (.mdc, .cursorrules)
Agent instructions
Project context files (AGENTS.md, CLAUDE.md)
Skills
On-demand workflows agents load when the task matches (SKILL.md)
Prompts
Reusable one-shot prompts for any model
Have a file you actually use?
Original submissions are listed as soon as they pass the scan. We are not ingesting vendor-official Anthropic or OpenAI files until the license is unambiguous.
Submit for scanning