← ClaudeAtlas

machine-forensics-looplisted

Diagnose and fix problems on any machine using a persistent, per-machine forensics knowledge base that the agent builds and maintains itself. Agent-agnostic - works for any coding agent, any OS. Use when the user reports something broken, failing, slow, missing, or asks to "check" a machine, an app, or an agent itself, before exploring from scratch. Reads this machine's existing knowledge base first, learns from recorded past mistakes, then attributes, reproduces, fixes, and writes findings back.
timmyjoe917/machine-forensics-loop · ★ 1 · AI & Automation · score 72
Install: claude install-skill timmyjoe917/machine-forensics-loop
# Machine Forensics Loop A discipline for diagnosing machine-level problems **without repeating past mistakes**. This package is the **manual**. It is identical for everyone, and it contains **no facts about any particular machine**. The **knowledge base** — one machine's own case notes — is built and maintained *on the machine*, by following the loop below. A machine that has never run this loop starts with an empty one. > **Manual ≠ case notes.** Reading someone else's knowledge base tells you how a *different* > machine broke. It is never evidence about yours. ## Who this is for **Any agent, on any machine — and the humans who work with them.** Not tied to a product, harness, or vendor. It applies whether you are Codex, another coding agent, or a custom harness of your own, and whether the machine runs Windows, macOS, or Linux. It earns its keep on a machine with **history**: installs, permission changes, environment variables, half-finished fixes. Most of that history is undocumented, and it is why the same fault costs full price twice. This skill writes it down, in the place the next agent will look. It needs only two things: 1. A place to keep the knowledge base (a plain folder is enough). 2. Willingness to follow the loop: **read first, verify with machine evidence, write back.** The two standing principles are deliberately agent-neutral: they are about how *any* agent should handle permissions, and how it should treat *any* other agent's conclusions — includ