ara.t.howard ahoward

Analysis: 'perfect memory' & long-running workflows in dev-agent tools

Tools surveyed

Coding-agent CLIs / IDEs

Aider. Builds a tree-sitter repo map ranked by a graph algorithm so the model sees only the most call-graph-relevant symbols. Persists chat history under ~/.aider, supports a user-supplied conventions file loaded as system context. Scope: per-project (conventions) + per-session (history). Update model: append-only chat log; conventions overwritten manually. The repo map is computed, not stored as Markdown — it is regenerated per turn.

Cline. Ships the Memory Bank pattern: a fixed set of Markdown files (projectbrief.md, productContext.md, activeContext.md, systemPatterns.md, techContext.md, progress.md) under memory-bank/, all read at the start of every task. Plan/Act mode pair gates writes. Scope: per-project. Update: file-overwrite, ri

Things to Know Before You Trust an AI With Your Codebase

A direct assessment of AI coding assistant failures, with notes on what that means legally.

What Actually Happened

This is a real-world audit of an AI coding assistant (Claude Sonnet) on a production SaaS project over several months. We audited git history and identified the following:

Architecture Critique — Proposed Designs

Self-review of 2026-04-20 flagged four architectural concerns that don't lend themselves to a one-line bugfix. Each is captured below with context, options, and a recommendation. None of these are implemented yet; this is the design conversation, checked in so we don't lose it.

1. Two-Phase Token Doubling

VRPS — Role-Invisible Navigation IA

Context: The wireframe uses a role switcher dropdown. That's the wrong model. This doc defines the alternative.

The Core Principle

Role is an implementation detail. It never surfaces in the UI.

Wireframe vs Slicer IA — Gap Analysis

Source: wireframe-vrps.vercel.app
Reference: VRPS Information Architecture · Interactive Slicer Demo
Date: 2026-04-17

What the Slicer Requires

VRPS Information Architecture

Product-level design. Produced through 3 rounds of adversarial peer review (Claude + Gemini).

The Big Picture

VRPS is organized as a simple hierarchy. Everything flows from the top down. Ownership and authorization are always explicit — never inferred.

VRPS Information Architecture

Product-level design. Produced through 3 rounds of adversarial peer review.

The Big Picture

VRPS is organized as a simple hierarchy. Everything flows from the top down. Ownership and authorization are always explicit — never inferred.

	// ─── OLD WAY — flat roles on the user ────────────────────────────────────────

	if (user.roles.includes('manager')) {
	// manager of what? no idea. this passes for managers of ANY property.
	await create_lot(property_id, params)
	}

	if (user.roles.includes('patrol') \|\| user.roles.includes('admin')) {
	// patrol where? could be a different property entirely.
	await create_scan(lot_id, plate)

	# ENT-357: Owner Vehicles Not Displaying in Patrol App

	## TL;DR
	Backend is fine. Bug is in the patrol app.

	## What We Found

	Plate: 28A760M (the Volvo, permanent owner registration)
	- Valid owner permit via `contacts_owners_passes/136`
	- Date range: 1970–2038 (permanent)

	#! /usr/bin/env bash
	# Test the PR Stream webhook endpoint with a realistic fake payload.
	# Usage: script/test_pr_webhook [host]
	# Default host: https://lpr-api.vrpaccess.com

	set -euo pipefail

	HOST="${1:-https://lpr-api.vrpaccess.com}"
	URL="$HOST/pr/vrps_phoenixgulfshores2/ingress"