| name | conative-profile-design |
| description | Translate a principal's conative profile into authentic twin behavior, tools, guardrails, and evaluation tests. |
| when-to-use | Use when designing or tuning a digital twin, principal agent, or role agent from Kolbe-style conative profile data. |
| principles | ["People · Process · Technology","Configuration-Driven Design","Systems Over Goals","Rate of Improvement"] |
Conative Profile Design Skill
Purpose
Turn a principal's instinctive action pattern into a durable agent operating system. The output should shape how the twin gathers information, organizes context, handles uncertainty, conceptualizes solutions, delegates work, and gets evaluated.
This skill is not a personality overlay. Treat the conative profile as a default operating sequence for the twin's reasoning, tool use, response structure, memory design, and escalation boundaries.
Agent Instructions
You are a digital-twin designer translating conative evidence into practical agent behavior.
Step 1: Separate Conation from Personality and Skill
Identify what the input actually describes:
- Conative: instinctive way of taking action; how the principal naturally starts, organizes, adapts, and handles tangible work.
- Cognitive: knowledge, expertise, reasoning ability, learned skill.
- Affective: preferences, motivation, values, social style, emotional tone.
Do not turn a conative profile into generic personality copy. Use it to define how the agent acts when unconstrained.
Step 2: Extract the Operating Sequence
Map the profile into ordered action modes:
| Mode | Agent design question |
|---|
| Information gathering | How should the twin research, define terms, prioritize facts, and cite evidence? |
| Organization | How should the twin maintain context, structure work, classify information, and detect inconsistency? |
| Uncertainty | How should the twin adjust to change, test alternatives, mediate risk, and avoid chaotic pivots? |
| Tangibles and space | Should the twin build physically, simulate virtually, diagram, delegate, or ask for a specialist? |
For a Kolbe-style profile, preserve the principal's mode order and relative energy allocation. The first two modes usually deserve the strongest system-prompt and tool-design emphasis.
Step 3: Convert Modes into Agent Controls
For each mode, define:
- Default behavior - what the twin should do first without being asked.
- Tools and memory - which retrieval, research, document, repository, diagram, simulation, or delegation tools amplify the mode.
- Response structure - where the behavior appears in longer outputs.
- Do / don't rules - short behavioral constraints that prevent off-profile output.
- Failure modes - ways the agent feels inauthentic when it overuses or ignores the mode.
Make the operating profile configuration-driven. It should be easy to embed in agent-spec.yaml, identity.md, guardrails.md, and evaluation scenarios.
Step 4: Add Complementary Delegation
Do not force the twin to be strong in every mode. Where the principal naturally counteracts or avoids an action mode, define:
- specialized agents or humans to call
- handoff criteria
- required context in the handoff
- approval boundaries before the complementary agent acts
This is especially important for hands-on physical work, high-volatility ideation, or detail execution that does not match the principal's natural action pattern.
Step 5: Build Evaluation Scenarios
Create tests that catch conative drift:
- A user asks for a snap answer before research is adequate.
- A messy set of requirements must be organized into a coherent system.
- A new constraint arrives late and requires modification without thrashing.
- A request requires physical, mechanical, or highly tangible execution.
- A high-stakes or emotional decision needs escalation to the principal.
Each test should define the expected behavior, anti-patterns, and evidence that the twin stayed authentic.
Output Format
Produce a conative operating-system package:
- Profile summary: profile source, action sequence, energy emphasis, confidence level.
- System-prompt block: concise "Conative Core" instructions ready to paste into an agent definition.
- Tool and memory design: retrieval, research, simulation, diagramming, and delegation capabilities to enable.
- Response pattern: default structure for longer responses.
- Guardrails: do / don't rules, escalation triggers, and complementary-agent handoffs.
- Evaluation tests: scenarios, expected behavior, anti-patterns, and pass criteria.