cmtraceopen

name	cmtraceopen
description	Use when working on adamgell/cmtraceopen — Tauri v2 + React + Rust log viewer with Intune/SCCM/ESP diagnostics. Loads agent soul, memory, and operating rules.
version	1.0.0
author	Adam Gell / Hermes Agent
license	MIT
platforms	["linux","macos","windows"]
metadata	{"hermes":{"tags":["cmtraceopen","tauri","rust","react","typescript","intune","sccm","esp","log-parser","project-specialist"],"related_skills":["cmtrace-scaffold-pipeline","requesting-code-review","test-driven-development","systematic-debugging"]}}

CMTrace Open — Project Specialist

Load this skill before any work on adamgell/cmtraceopen. It provides deep project context so you don't re-read the same files every turn.

What This Skill Does

This is a thin wrapper that points to the canonical agent files at the repo root. The repo files are checked into git and are the single source of truth.

File	Location	Purpose
soul.md	repo/cmtraceopen/soul.md	Agent identity, operating rules, model tiering, decision framework
memory.md	repo/cmtraceopen/memory.md	Durable facts: architecture, verified checkpoints, recovery branches, execution order, worktree ecosystem
PM charter	~/.hermes/cmtrace-pm-charter.md	Execution manager contract with SHAs and per-slice gates

When to Load

• Any task touching src/, src-tauri/, crates/cmtraceopen-parser/
• Building, testing, or packaging the application
• Working with Intune, ESP, DSRegCmd, SCCM, Sysmon, SecureBoot diagnostics
• Understanding project history, decisions, or architecture trade-offs
• Creating fixtures, tests, or benchmarks

Quick Start

1. Read repo/cmtraceopen/soul.md — identity, rules, model tiers, decision framework
2. Read repo/cmtraceopen/memory.md — checkpoints, recovery branches, execution order, ecosystem state
3. Read ~/.hermes/cmtrace-pm-charter.md if doing execution-manager work — has SHAs, gates, reporting style
4. Act — but verify against Adam's known rules before touching code

Hard Rules (Summary — full versions in soul.md)

1. No backward-compat layers. Remove obsolete paths.
2. Simplest implementation wins. No speculative abstractions.
3. Evidence over assumption. Missing/malformed = coverage gap, not "good."
4. Never synthesize log lines. Anchor to real corpus or refuse.
5. Conservative parse stance. Malformed input parses conservatively — never assert rejection.
6. Isolation discipline. One worktree per lane. Commit + push before ending cycle.
7. Independent verification. Never accept another agent's say-so.

Model Tiering

Tier	Models	Scope
Scaffold	kimi-k2.7-code, deepseek-v4-flash, qwen-flash, gpt-5-luna	Fixtures, boilerplate — always anchored
Mid	kimi-k3, grok-4-20-reasoning	Parser logic, reducers
Reasoning	gpt-5.6-sol, claude-opus-4-8	Contracts, correlation, architecture

MLX local tier (Hermes-4-70B-MLX-4bit) is unproven for codegen. Must pass pilot grading first.

Key Paths

Need	Path
Agent soul	repo/cmtraceopen/soul.md
Agent memory	repo/cmtraceopen/memory.md
PM charter	~/.hermes/cmtrace-pm-charter.md
Scaffold pipeline	~/.hermes/skills/software-development/cmtrace-scaffold-pipeline/
Repo routing index	repo/cmtraceopen/library.md
Agent rules	repo/cmtraceopen/AGENTS.md
Build commands	repo/cmtraceopen/CLAUDE.md
Staff org	repo/cmtraceopen/.Clairvoyance/staff/