Michael Byrne
Search

Free-Public-Tools

Free tools for AI assistants and coding agents, a new one every week. Each started in my own work with AI and is tested before release.

By

Published
Role
Author: each tool, its tests and its pages, built with AI assistance
Status
active
Source on GitHub
Summary
Five tools published, each with its own page, notes and every version, and one published test with all its marks.

Context

Free-Public-Tools are free tools for AI assistants and coding agents, released one a week. Each one started in my own work with AI and is tested before release: instructions, commands, subagents and small checkers that I use myself. Use, copy, change and share them however you like, with no need to credit anyone (MIT-0).

What is in it

Each tool has its own page. Copy it in one tap and paste it where it goes: in a chat, in a chat app’s custom instructions, a CLAUDE.md or AGENTS.md file, or your coding agent. A tool made of several files copies as an install prompt that fetches the folder for you. The first five:

  • Free-Public-Tools card for Epistemic stance: “Make your AI push back.” You: “This change doubled our sign-ups!” Your AI, with Epistemic stance: “Maybe. Sign-ups jumped last March too. Let’s rule that out first.”

    Epistemic stance: make your AI push back with evidence instead of agreeing, and say how sure it is about the claims that matter.

  • Free-Public-Tools card for Adversarial self-review: “Make your AI show the proof before it says done.” You: “Is the login fix done?” Your AI, with Adversarial self-review: “Not yet. I undid my fix on purpose, and the tests still passed.”

    Adversarial self-review: a command that has your coding agent check its own change before calling it done, with pasted evidence for each check.

  • Free-Public-Tools card for Claim auditor: “Check your AI’s numbers before you publish.” You: “Check my findings with the claim auditor.” Your AI, with Claim auditor: “Three claims check out. One doesn’t: your data shows 24% faster, not 40%.”

    Claim auditor: a subagent that checks every claim in a README, changelog or benchmark write up before you post it, and grades each one.

  • Free-Public-Tools card for Decision challenge: “Challenge a decision before you lock it in.” You: “Let’s check for new payments every 5 minutes.” Your AI, with Decision challenge: “Before we lock it in, let’s compare it with two real alternatives.”

    Decision challenge: a command that pressure tests one decision before it becomes hard to reverse.

  • Free-Public-Tools card for Agent assets lint: “Keep your AI’s instruction files in shape.” You: “Is my CLAUDE.md getting too long?” Your AI, with Agent assets lint: “Yes: 240 lines against your 200 limit, and one skill has no description.”

    Agent assets lint: a small Node tool that checks instruction files, skills and agent files against rules you write in one JSON file.

Evidence

Everything is public: the tools site, and the repository, where every tool’s folder holds the tool, a README with what it does and where it goes, and all its versions.

The first published test put Decision challenge against two plain prompts on 18 decisions from six of my own projects, across seven AI models: 378 runs on 14 October 2026. Answer keys were written before the runs, and replies were marked without the markers knowing which version or model wrote them. No message asked for an edit, yet 21 of 126 runs with the short prompt and 43 with the plain request edited or deleted project files; with the command, none did. In real use it got 70% of decisions right, against 60% and 51%. With six projects and one run per model and question, a gap of a few points is noise, and I wrote the questions myself. Every question and every mark is on the test page.

Current state

Five tools are out and the series adds one a week. Found a problem, or a tool that does not behave the way its README says? Open an issue.

A line drawing of a band on a break: stick-figure musicians resting on an amp, at a drum kit and on the floor with drinks, chatting and stretching among a sitar, a tuba and a harp, under a rail of hanging guitars, banjos, chimes and a trumpet.