Ian Maurer imaurer

name	adversarial-project-review
description	Review a whole project against its own claims, in order of harm: correctness, security and data rights, release, gate trust, reliability, performance and maintainability. A fresh reviewer runs it read-only before a release, at each milestone's end, and when the project stalls, and the PM turns each accepted finding into a ticket.

Adversarial project review

Review the project as one system, and try to break its promises. Look hardest where two parts meet: code and spec, parent and child process, trusted and untrusted input, a record's start and its end, an idle machine and a loaded one, old data and a migration, the tested build and the shipped one, and a ticket's intent and what its acceptance can prove.

Who runs it and when

pm: a Git-only work tracker for a fleet of coding agents

pm is a command-line tool that keeps tickets, plans, decisions, ideas, and mail between agents as plain Markdown files in Git, runs builds on a shared build machine, and shows a team of coding agents what to do next. This document describes its behaviour and file formats closely enough that a coding agent can rebuild it in any language.

1. Why it exists

The author is a CTO who dictates work to a fleet of coding agents. The agents run in tmux sessions on one coordination machine, across several projects at once, and their builds and tests run on a separate build machine. Each project has a planning agent and a build agent, and those agents start subagents to write and review code.

That setup broke in predictable ways:

description	Build tests outside-in: describe behavior first, grow tests red-green, keep what states behavior, and remove the scaffolding before landing.

Outside-in tests

Test-driven development works well for agents. The trouble comes after green. Each step leaves a small test behind, and the pile grows into test sediment. Sediment slows every build and change. It breaks on refactors that keep behavior. It can lock in wrong behavior, so a bug fix means defending the old wrong answer first.

OpenClaw's test-audit skill cleans up a pile that already exists. Use it for audits. This skill keeps the pile from forming: describe behavior from the outside, grow tests red-green, keep what states behavior, and remove the rest before landing.

The goal

description	Notice, diagnose, address and prevent the proof spiral: agents spending more effort proving, recording and defending work than doing it.

Proof spiral

A proof spiral is work about the work growing faster than the work. A check needs a record, the record a review. The review finds a gap in the check, so the agent builds a better checker. Each step looks reasonable. Together they stall delivery while the logs look rigorous. Inspect the workflow first. Rules often make defects and violations explicit but leave delivery time and acceptable leftover uncertainty unstated. Strong instruction following then becomes a liability.

1. Notice

Strongest signs:

Unsloth Gemma 4 MOE Setup Guide

Running a 26B parameter model with only 6GB RAM using mmap

Overview

This guide covers running Gemma 4 26B MOE (Mixture of Experts) locally on an Apple Silicon Mac using llama.cpp with memory-mapped files. The MOE architecture activates only 8 of 128 experts per token, making it incredibly efficient.

Spec	Value

A synthesis of Andrej Karpathy's No Priors interview (March 2026), the auto research ecosystem, and what it all means for how we build things.

The Shift: Manifesting, Not Coding

[0:00–6:36]

Karpathy says he hasn't typed a line of code since December 2025. Not because he stopped working — he works 16 hours a day — but because the verb changed:

Task

Launch a new agent that has access to the following tools: Bash, Glob, Grep, LS, exit_plan_mode, Read, Edit, MultiEdit, Write, NotebookRead, NotebookEdit, WebFetch, TodoRead, TodoWrite, WebSearch. When you are searching for a keyword or file and are not confident that you will find the right match in the first few tries, use the Agent tool to perform the search for you.

When to use the Agent tool:

If you are searching for a keyword like "config" or "logger", or for questions like "which file does X?", the Agent tool is strongly recommended

When NOT to use the Agent tool:

If you want to read a specific file path, use the Read or Glob tool instead of the Agent tool, to find the match more quickly
If you are searching for a specific class definition like "class Foo", use the Glob tool instead, to find the match more quickly
If you are searching for code within a specific file or set of 2-3 files, use the Read tool instead of the Agent tool, to find the match more quickly

prompt chatgpt 5.2 pro (12/31/2025)

How would you improve the prompt based on the below document? Be explicit, provide detailed analysis of what's missing, what could be better, what would be tweaked. And then after all your analysis and bullet points, give me the new prompt.

original skill

https://github.com/steipete/agent-scripts/blob/main/skills/create-cli/SKILL.md

document

Determining the Minimum Supported Pydantic Version (Binary Cut)

This note documents how we determined the lowest Pydantic 2.x version that works across our supported Python versions (3.10–3.14) using uv and pytest. It also records the final requirement change made to pyproject.toml.

Objective

Identify the minimum Pydantic version that passes our test suite across Python 3.10, 3.11, 3.12, 3.13, and 3.14.
Update our dependency constraint accordingly.

	# File: run.py

	#!/usr/bin/env python3
	"""
	Working demonstration of the BioMCP API with actual function calls.
	This shows the corrected API in action.
	"""

	import asyncio
	from biomcp.variants.search import (