ctrl-alt-pray
Engineering Publications

Autonomous AI Agent Autopsy & Field Manuals

Hard architectural analyses, process tree forensics, and epistemic debugging methodologies for developers building with Cursor, Claude Code, and autonomous coding agents.

SYSTEMS 6 min read

Why Autonomous AI Coding Agents Enter Doom Loops (And Why Prompts Can't Stop Them)

Dissecting the autoregressive context contamination and cognitive tunnel vision that causes agents to churn dirty files in a 3:00 AM loop, and why deterministic circuit breakers outperform prompt pleas.

OS & PROCESSES 8 min read

Killing Ghost Terminals: How to Stop AI Agents from Freezing on Port Conflicts and Prompts

The forensics behind silent terminal hangs: unhandled (y/n)? stdin prompts, rogue watch-mode loops, and orphaned worker threads locking ports :3000 and :5173. Introducing the Cascade Tree Killer.

METHODOLOGY 10 min read

Epistemic Debugging: 12 Canonical Falsification Probes to Force AI Agents into Reality

Applying Karl Popper's falsificationism to autonomous coding. Replacing polite conversational apologies with the 12 Canonical Rites: Wrong Altar, Check the Check, and Clean-Context Resurrection.

TUTORIAL 4 min read

Arming Cursor, Claude Code, and Windsurf: The 10-Second MCP Safety Net Guide

A hands-on, zero-friction guide to provisioning the 2-Strikes Circuit Breaker across all 9 major editors, checking visual telemetry on port 3900, and saving tokens with zero runtime dependencies.