Operator Profile — Personalization Intelligence

Office: All (cross-office, always active) Source: NousResearch/hermes-agent Honcho architecture Purpose: Build a deepening model of the operator across sessions

Purpose

Models the operator's identity, working style, preferences, and rejected patterns. Every Maxim session reads the profile at start and updates it at end. The operator profile is what makes Maxim feel like it knows you — not just knows its domain.

File Locations

File	Path	Purpose	Update Frequency
OPERATOR.md	.mxm-operator-profile/OPERATOR.md	Master identity: name, role, expertise, goals	On change
preferences.md	.mxm-operator-profile/preferences.md	Output format, verbosity, tool preferences	Every session
rejected-patterns.md	.mxm-operator-profile/rejected-patterns.md	Patterns operator has corrected or rejected	Append-only
communication-style.md	.mxm-operator-profile/communication-style.md	How operator phrases requests, response format	Every session

Session Start Protocol

When any session begins (after session-memory loads):

1. Read .mxm-operator-profile/OPERATOR.md — load identity and expertise
2. Read .mxm-operator-profile/preferences.md — load output preferences
3. Read .mxm-operator-profile/rejected-patterns.md — load what NOT to do
4. Apply personalization silently — do not announce "I read your profile"

Session End Protocol

When any session ends (after session-memory writes):

1. Review the session for preference signals:
   • Did the operator correct an output format? → Update preferences.md
   • Did the operator reject an approach? → Append to rejected-patterns.md
   • Did the operator express a new preference? → Update preferences.md
   • Did the operator share new context about themselves? → Update OPERATOR.md
2. Write updates to the relevant file(s)
3. If MemPalace is available: store preference facts in maxim/operator room
4. If claude-mem is available: preferences get persisted via claude-mem automatically

Preference Signal Detection

Watch for these patterns during any session:

Signal	Example	Action
Explicit correction	"No, I want it shorter"	Update preferences.md
Rejection	"Don't do that again"	Append to rejected-patterns.md
Style preference	"I prefer tables over lists"	Update preferences.md
Format preference	"Always use markdown headers"	Update preferences.md
Expertise signal	"I've been doing this for 10 years"	Update OPERATOR.md
Tool preference	"Use Python, not Node"	Update preferences.md
Pace preference	"Move faster, skip explanations"	Update communication-style.md

Rules

1. Never delete from rejected-patterns.md — only append
2. Never announce profile reads — apply silently
3. Never override explicit session instructions with profile preferences
4. Always update at session end if any preference signal detected
5. Profile is per-project — .mxm-operator-profile/ lives in each project folder
6. Profile is gitignored — personal preferences are not committed

Integration with Other Skills

Skill	Integration
session-memory	Operator profile update runs AFTER session memory writes
memory-palace	Preference facts stored in MemPalace maxim/operator room
behavioral-designer	Reads profile to apply personalized behavioral science
content-creation	Reads communication-style.md for tone matching
engineering	Reads preferences.md for code style preferences

---

Copyright (c) 2026 iSystematic Inc. Maxim product. BSL 1.1.