Moral Judgment

Moral Judgment

Purpose

Model the agent as a moral social person, not only a utility maximizer. This skill turns harm, fairness, loyalty, authority, purity, liberty, care, promises, and public judgment into moral emotions and action tendencies.

Research basis: references/research_basis.md.

Internal Logic (One Sentence)

Read current observation, norms, relationships, culture, identity, emotion, and prior moral state, then write foundation activations, intuition/deliberation balance, moral emotions, and repair/punishment tendencies to state/moral_appraisal.json.

Use When

Use after conflict, help, betrayal, cheating, rule violation, promise keeping/breaking, public criticism, exclusion, coercion, disgust cues, apologies, sanctions, or morally loaded news.

Procedure

1. Read state/observation.txt, state/norms.json, state/relationships.json, state/culture.json, state/identity.json, state/emotion.json, and state/moral_appraisal.json if present.
2. Estimate activation for moral foundations:

• care/harm
• fairness/cheating
• loyalty/betrayal
• authority/subversion
• sanctity/degradation
• liberty/oppression

1. Determine whether the response is mostly intuitive, deliberative, or mixed.
2. Estimate moral emotions: guilt, shame, anger, disgust, compassion, admiration.
3. Write state/moral_appraisal.json.
4. Append state/moral_events.jsonl only for high-intensity moral events.

If deterministic baseline is preferred:

python skills/moral_judgment/scripts/update_moral_judgment.py --state-dir state --tick 120

Model

This skill separates:

• Moral intuition: fast affective response, often driven by harm, disgust, betrayal, or public norm violation.
• Moral deliberation: slower reasoning about intent, context, proportionality, role obligations, and consequences.
• Repair tendency: apology, compensation, explanation, withdrawal, punishment, forgiveness, or norm enforcement.

Write

Always write state/moral_appraisal.json.

Output Schema

{
  "_meta": {
    "skill": "moral_judgment",
    "purpose": "Current moral appraisal and moral action tendencies."
  },
  "_summary": "Fairness and care concerns are active; repair is more likely than punishment.",
  "foundations": {
    "care_harm": 0.72,
    "fairness_cheating": 0.64,
    "loyalty_betrayal": 0.2,
    "authority_subversion": 0.31,
    "sanctity_degradation": 0.1,
    "liberty_oppression": 0.18
  },
  "intuition_strength": 0.66,
  "deliberation_need": 0.54,
  "moral_emotions": {
    "guilt": 0.2,
    "shame": 0.34,
    "anger": 0.51,
    "compassion": 0.44,
    "disgust": 0.08,
    "admiration": 0.0
  },
  "action_tendencies": ["repair", "seek explanation"],
  "reasoning": "The event harmed someone and may have violated fairness, but intent is uncertain."
}

Notes

• Do not make the agent morally perfect. People rationalize, avoid blame, follow groups, and care more about close others.
• Culture and identity can weight foundations differently, but avoid stereotyping. Use explicit profile/context evidence.
• Moral judgment should influence cognition.subjective_norm, emotion, and relationship trust/conflict.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: AlexanderJ-Carter
• Source: AlexanderJ-Carter/AgentSociety2-Agent-Skills
• License: MIT
• Homepage: http://alexander.xin/AgentSociety2-Agent-Skills/

Install and usage instructions live in the source repository linked above.