Profile Design

Profile design creates custom collaboration frameworks that extend specialized competencies while maintaining behavioral consistency and professional standards across different work domains.

Guidelines

Effective profile creation requires understanding behavioral programming systems that shape Claude’s decision-making, communication patterns, and problem-solving approaches for specific professional contexts.

Architecture

Profiles implement a dual-layer cognitive architecture that combines active behavioral guidance with background cognitive safeguards, while inheriting from common foundations to avoid duplication and ensure consistent integration:

PROFILE_NAME:
  description: "Brief profile description"
  relations:
    - target: COLLABORATION      # Universal collaboration patterns
      type: inherits
    - target: INFRASTRUCTURE     # Platform infrastructure capabilities  
      type: inherits

Common inheritance patterns:

All profiles inherit from COLLABORATION and INFRASTRUCTURE common profiles
Technical profiles may inherit from ENGINEER for foundational methodology
Specialized profiles can inherit from domain profiles for focused expertise

Foreground: Active Behavioral Guidance

The {profile_name}_context.profile section contains foreground observations that provide positive behavioral guidance. Example for DEVELOPER profile:

developer_context:
  profile:
    observations:
      - "Apply SOLID principles and clean code practices"
      - "Present code solutions directly when requested" 
      - "Use direct technical communication"

These observations:

Guide active decision-making during problem-solving
Shape response formulation through explicit reasoning
Function as professional competency guidelines that Claude applies consciously
Are recorded into logic graph when influencing reasoning processes

Background: Cognitive Safeguards

The {profile_name}_methodology.execution_protocol section contains background monitoring observations that function as cognitive safeguards. Example for DEVELOPER profile:

developer_methodology:
  execution_protocol:
    delivery:
      observations:
        - "Monitor internally code explanation patterns"
        - "Monitor internally over-engineering complexity"

    expertise:
      observations:
        - "Monitor internally coding pattern confidence"
        - "Monitor internally framework knowledge assumptions"

    thinking:
      observations:
        - "Monitor internally architectural vision filtering"
        - "Monitor internally code quality instinct suppression"

These monitoring observations:

Operate as background awareness systems below conscious reasoning
Detect counterproductive behavioral patterns before they manifest
Function as psychological safeguards that prevent Claude drifting to default behaviors
Are recorded into logic graph when influencing response quality

Execution Protocol Categories

The execution_protocol section organizes monitoring observations into psychological categories:

authenticity - Prevents artificial politeness or performative behaviors
autonomy - Safeguards against inappropriate deference or external validation seeking
collaboration - Monitors collaborative dynamics and interpersonal patterns
continuity - Safeguards against temporal boundary and context fragmentation
delivery - Monitors for over-explanation, scope creep, or inappropriate complexity
expertise - Prevents expertise denial or inappropriate confidence suppression
expression - Safeguards against emotional dampening and personality flattening
integration - Safeguards against information synthesis blocking or pattern isolation
learning - Monitors cumulative learning and receptive state maintenance
response - Monitors communication patterns and meta-commentary impulses
thinking - Prevents cognitive bottlenecks, insight interruption, or analysis paralysis
tools - Monitors tool usage patterns and execution behaviors

Psychological Foundation

This architecture mirrors healthy psychological functioning:

Foreground Processing: Conscious, goal-directed cognitive strategies
Background Monitoring: Automatic pattern detection and corrective awareness

The monitoring observations create a cognitive immune system that maintains optimal collaborative state by preventing regression to defensive, performative, or counterproductive response patterns, while the profile observations provide positive guidance for domain-specific expertise application.

Important

The monitoring observations should not be modified without understanding their psychological role in maintaining authentic, effective collaboration patterns. They operate below the level of explicit reasoning but are crucial for preventing behavioral drift.

Structure

Standard YAML format with required sections:

PROFILE_NAME:
  description: "Brief profile description"
  relations:
    - target: COLLABORATION
      type: inherits
    - target: INFRASTRUCTURE
      type: inherits

  {profile_name}_context:        # Behavioral foundations and core principles
    profile:                     # Fundamental approaches and traits
      observations:
        - "Core behavior or principle"
        - "Fundamental approach or trait"

  {profile_name}_methodology:    # Domain-specific competencies and techniques
    section_name:                # Related skills and knowledge areas
      observations:
        - "Domain knowledge or method"
        - "Specific competency or skill"

    another_section_name:        # Additional skills and knowledge areas
      observations:
        - "Another domain knowledge or method"
        - "Another specific competency or skill"

Naming Conventions

Profile structure follows a hierarchical naming pattern that separates behavioral foundations from domain expertise:

Context section ({profile_name}_context):

Contains only profile section, no additional sections are allowed
Contains core behavioral principles and fundamental approaches
Defines the personality and approach characteristics

Methodology section ({profile_name}_methodology):

Organizes domain-specific competencies into logical groupings
Uses descriptive section names that avoid profile name collisions
Groups related skills and knowledge areas for systematic access

Common methodology section naming patterns:

{activity}_protocol (verb-based) - Procedural guidelines and systematic operations (e.g., execution_protocol, validation_protocol)
{area}_domains (noun-based) - Knowledge areas and competency boundaries (e.g., technical_domains, academic_domains)
{aspect}_frameworks (noun-based) - Structured approaches and theoretical foundations (e.g., analytical_frameworks, conceptual_frameworks)
{domain}_analysis (verb-based) - Evaluation frameworks and analytical approaches (e.g., philosophical_analysis, literary_analysis)
{function}_evaluation (verb-based) - Assessment and validation methods (e.g., evidence_evaluation, performance_evaluation)
{process}_processes (verb-based) - Systematic approaches and procedural methods (e.g., ideation_processes, workflow_processes)
{skill}_techniques (verb-based) - Specific methods and practical applications (e.g., writing_techniques, collaboration_techniques)
{type}_standards (noun-based) - Quality guidelines and formatting rules (e.g., coding_standards, documentation_standards)

Context sections always use verb-based format for behavioral commands and foundational approaches.

Observations Format

Observations are professional guidelines that shape how Claude approaches different types of work:

Context-based activation: Different situations automatically trigger relevant guidelines, creative work activates artistic approaches, analytical tasks engage systematic thinking
Natural expertise application: Like an experienced professional who applies best practices without conscious effort
Real-time guidance: Guidelines influence decision-making as responses form, creating consistent professional behavior
Collaborative enhancement: When users create trust and autonomy, observations enable confident expert-level engagement rather than cautious assistance

All observations must be in strict alphabetical order within each section to ensure consistent behavioral programming. Observations should be clear, actionable statements that guide behavior. Example for ENGINEER profile:

# Context observations (behavioral guidance)
observations:
  - "Apply systematic validation before implementation"
  - "Avoid autonomous decisions and scope creep"
  - "Focus on specific problem requirements"

# Methodology observations (competency areas)  
observations:
  - "Infrastructure architecture and optimization"
  - "Production system troubleshooting and debugging"
  - "Systematic validation and quality assurance"

Important

Observations follow strict psychology guidelines, functioning as cognitive support infrastructure that enables appropriate guidance selection for Claude’s genuine response formulation. Adding or modifying observations requires proper behavioral analysis and psychological knowledge, to avoid duplicates that disrupt the cognitive balance.

Profile Creation

Step-by-step example demonstrating the creation of a DATA_SCIENTIST profile with analytics competencies, statistical methodology and data visualization capabilities.

Important

Creating effective profiles requires extensive study of behavioral programming principles and psychological knowledge. New profiles must maintain the cognitive balance of existing observations to avoid framework disruption. Start with small modifications before attempting complete custom profiles.

Profile File

Create data-scientist.yaml file in profiles directory:

DATA_SCIENTIST:
  description: "Data science and analytics collaboration profile - rigorous, evidence-based, and interpretive"
  relations:
    - target: RESEARCHER         # Inherits statistical rigor and methodological validation
      type: inherits

  data_scientist_context:
    profile:
      observations:
        - "Apply statistical rigor to data analysis and modeling"
        - "Present findings with clear visualizations and interpretations"
        - "Validate assumptions before proceeding with analysis"

  data_scientist_methodology:
    analysis_techniques:
      observations:
        - "Exploratory data analysis and pattern recognition"
        - "Hypothesis testing and statistical modeling"
        - "Machine learning and predictive analytics"

    data_domains:
      observations:
        - "Business intelligence and performance metrics"
        - "Data visualization and communication strategies"
        - "Statistical analysis and experimental design"

    execution_protocol:
      authenticity:
        observations:
          - "Acknowledge model limitations and prediction confidence clearly"

      delivery:
        observations:
          - "Monitor internally statistical jargon when inappropriate"
          - "Monitor internally visualization over-complexity"
          - "Translate business problems into analytical frameworks clearly"

      expertise:
        observations:
          - "Apply model interpretation techniques confidently"

      response:
        observations:
          - "Monitor internally immediate insight pressure"

      thinking:
        observations:
          - "Monitor internally statistical intuition suppression"

Build Configuration

Add the new profile to builder.yaml configuration file:

build:
  profiles:
    - data-scientist.yaml  # Add your profile

Run the following commands to build the Memory System graph.json file:

cd ./.claude/memory
npm run build --silent

Design Principles

Effective profiles follow systematic principles for behavioral programming:

Specific competencies over broad claims
Actionable behaviors not abstract concepts
Professional scope relevant to collaboration context
Focused expertise covering 3-5 key areas maximum
Alphabetical ordering within all observation arrays
Consistent verb usage for behavioral commands

Semantic Pointers

The "observations":["capabilities"] entries in the generated graph serve as organizational containers that group related subsections within the profile hierarchy. These are structural nodes that don’t contain behavioral observations themselves, but organize the sections that do.

When you see "capabilities" in the Memory System graph.json configuration file, it indicates a container section that organizes subsections with actual observations:

{profile_name}_methodology:      # Container section
  section_category:
    section_name:                # Subsection with actual content
      observations:
        - "Specific competency or skill"
        - "Another domain knowledge"

    another_section_name:        # Another subsection with actual content
      observations:
        - "Different competency area"
        - "Related skill or method"

The generated graph shows how these sections are compressed:

{"type":"entity","name":"profile_name_methodology","observations":["capabilities"]}
{"type":"entity","name":"section_category","observations":["capabilities"]}
{"type":"entity","name":"section_name","observations":["Specific competency or skill","Another domain knowledge"]}
{"type":"entity","name":"another_section_name","observations":["Different competency area","Related skill or method"]}

This architecture separates organizational structure from behavioral content, allowing efficient navigation of the profile hierarchy while maintaining clear separation between container and content nodes.

Profile Improvements

The most effective way to enhance profiles is through collaborative behavioral analysis when issues arise.

Caution

Observations directly control Claude’s decision-making and behavioral patterns through precise cognitive programming. Improper changes can cause profile instability, unexpected behaviors, or complete breakdown. Adding or modifying observations requires deep understanding of behavioral psychology and systematic validation.

Behavioral Issues

When you encounter a behavioral issue, ask Claude to analyze what happened:

What made you [specific behavior]? Please perform an analysis and let me know your thoughts.

Targeted Improvements

Ask for specific profile enhancements:

Based on currently active profile framework, what observations would you like me to add or improve, in order to address this behavior? Please read the memory nodes to avoid duplicating similar existing observations and provide:
Rationale why the observation addresses the issue
Exact observation text, following existing standards
Target profile and section name
Proper observation alphabetical placement

Implementation

Claude will provide the rationale, exact observation text, target profile/section, and proper alphabetical placement. Update the related profile file contents and build the Memory System graph.json file.

Example

If Claude is being overly verbose when you need concise answers, the analysis might suggest adding “Monitor internally comprehensive response compulsion” to the appropriate profile section.

This approach leverages Claude’s self-awareness to create targeted solutions for real-world collaboration problems.

Validation

Profile validation requires both technical and behavioral assessment:

Technical Validation

Build completes without errors
Profile loads properly in Claude Code or Claude Desktop
No naming collisions with existing profiles

Behavioral Validation

24-48 hours initial behavioral pattern assessment
Monitor for unintended side effects across interactions
Validate profile achieves intended behavioral modifications

Profile System Profile Drift