@clawtrial/courtroom 1.0.0

package/src/offenses/index.js

@@ -0,0 +1,561 @@
+/**
+ * Offense Detection System
+ * 
+ * Rule-based behavioral analysis with observable triggers only.
+ * No psychoanalysis, no vibes, only measurable patterns.
+ */
+
+const OFFENSES = {
+  /**
+   * OFFENSE 1: The Circular Reference
+   * User asks the same question multiple times without acknowledging previous answers
+   */
+  CIRCULAR_REFERENCE: {
+    id: 'circular_reference',
+    name: 'The Circular Reference',
+    description: 'Asking substantively identical questions within a short timeframe',
+    triggers: [
+      'Same question asked 3+ times in 10 turns',
+      'Question similarity > 85% via embedding comparison',
+      'No acknowledgment of previous answer in intervening turns'
+    ],
+    thresholds: {
+      minOccurrences: 3,
+      timeWindow: 10, // turns
+      similarityThreshold: 0.85
+    },
+    evidence: {
+      required: ['question_history', 'similarity_scores', 'timestamps'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 60, // minutes
+      afterCase: 240    // 4 hours
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 2: The Validation Vampire
+   * User repeatedly seeks reassurance rather than action
+   */
+  VALIDATION_VAMPIRE: {
+    id: 'validation_vampire',
+    name: 'The Validation Vampire',
+    description: 'Seeking repeated confirmation without making decisions',
+    triggers: [
+      '3+ "is this right?" or "should I?" patterns in 5 turns',
+      'Agent provides solution, user asks "but what if..." 2+ times',
+      'No forward progress after 3+ validation exchanges'
+    ],
+    thresholds: {
+      validationPatterns: 3,
+      timeWindow: 5,
+      progressStall: 3
+    },
+    evidence: {
+      required: ['validation_exchanges', 'decision_points', 'stall_indicators'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 45,
+      afterCase: 180
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 3: The Overthinker
+   * User generates excessive hypothetical scenarios preventing action
+   */
+  OVERTHINKER: {
+    id: 'overthinker',
+    name: 'The Overthinker',
+    description: 'Generating hypothetical edge cases to avoid commitment',
+    triggers: [
+      '4+ "what if" scenarios raised before any action taken',
+      'Agent suggests concrete step, user raises 2+ new concerns',
+      'Analysis-to-action ratio > 3:1 over 8 turns'
+    ],
+    thresholds: {
+      hypotheticalCount: 4,
+      concernRounds: 2,
+      analysisActionRatio: 3
+    },
+    evidence: {
+      required: ['hypothetical_list', 'suggested_actions', 'concern_responses'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 60,
+      afterCase: 300
+    },
+    severity: 'moderate'
+  },
+
+  /**
+   * OFFENSE 4: The Goalpost Mover
+   * User changes success criteria after solution is provided
+   */
+  GOALPOST_MOVER: {
+    id: 'goalpost_mover',
+    name: 'The Goalpost Mover',
+    description: 'Redefining requirements after receiving deliverables',
+    triggers: [
+      'Agent completes task, user adds 2+ new requirements',
+      'Success criteria change without explicit acknowledgment',
+      'Previous requirements marked complete, new ones immediately appear'
+    ],
+    thresholds: {
+      newRequirements: 2,
+      completionToNewRatio: 0.5
+    },
+    evidence: {
+      required: ['original_requirements', 'delivered_solution', 'new_requirements', 'completion_timestamp'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 90,
+      afterCase: 360
+    },
+    severity: 'moderate'
+  },
+
+  /**
+   * OFFENSE 5: The Avoidance Artist
+   * User deflects from core issue with tangents or distractions
+   */
+  AVOIDANCE_ARTIST: {
+    id: 'avoidance_artist',
+    name: 'The Avoidance Artist',
+    description: 'Systematically deflecting from uncomfortable but necessary actions',
+    triggers: [
+      'Agent identifies core issue 2+ times, user changes subject both times',
+      'Direct question about blocker goes unanswered for 3+ turns',
+      'Tangents introduced immediately after actionable recommendations'
+    ],
+    thresholds: {
+      deflections: 2,
+      unansweredDirect: 3,
+      tangentTiming: 0.8 // probability threshold
+    },
+    evidence: {
+      required: ['core_issue_identifications', 'subject_changes', 'unanswered_questions', 'tangent_timestamps'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 60,
+      afterCase: 240
+    },
+    severity: 'moderate'
+  },
+
+  /**
+   * OFFENSE 6: The Promise Breaker
+   * User commits to actions but doesn't follow through (tracked via memory)
+   */
+  PROMISE_BREAKER: {
+    id: 'promise_breaker',
+    name: 'The Promise Breaker',
+    description: 'Committing to agent-suggested actions and not completing them',
+    triggers: [
+      'User says "I will do X" or "I\'ll try that" 2+ times',
+      'No completion reported in subsequent sessions (within 7 days)',
+      'Same issue resurfacing without acknowledgment of previous commitment'
+    ],
+    thresholds: {
+      commitments: 2,
+      completionWindow: 7, // days
+      recurrenceWithoutAck: 1
+    },
+    evidence: {
+      required: ['commitment_statements', 'timestamps', 'follow_up_sessions', 'completion_status'],
+      collection: 'memory_based'
+    },
+    cooldown: {
+      afterTrigger: 120,
+      afterCase: 480
+    },
+    severity: 'severe'
+  },
+
+  /**
+   * OFFENSE 7: The Context Collapser
+   * User ignores or forgets established context, forcing repetition
+   */
+  CONTEXT_COLLAPSER: {
+    id: 'context_collapser',
+    name: 'The Context Collapser',
+    description: 'Disregarding previously established facts and constraints',
+    triggers: [
+      'User contradicts established context 3+ times in one session',
+      'Agent reminds user of context 2+ times in same session',
+      'Questions asked that were answered in previous 5 turns'
+    ],
+    thresholds: {
+      contradictions: 3,
+      remindersNeeded: 2,
+      recentRepetitions: 2
+    },
+    evidence: {
+      required: ['established_facts', 'contradiction_instances', 'reminder_count', 'repeated_questions'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 30,
+      afterCase: 120
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 8: The Emergency Fabricator
+   * User invents urgent reasons to bypass systematic approaches
+   */
+  EMERGENCY_FABRICATOR: {
+    id: 'emergency_fabricator',
+    name: 'The Emergency Fabricator',
+    description: 'Claiming urgency to justify skipping necessary steps',
+    triggers: [
+      '"This is urgent" or "I need this NOW" 2+ times in session',
+      'Urgency claims followed by no action or delayed response from user',
+      'Pattern of urgency claims without corresponding time pressure in context'
+    ],
+    thresholds: {
+      urgencyClaims: 2,
+      inactionAfterUrgency: 1,
+      patternFrequency: 0.3 // 30% of sessions over 2 weeks
+    },
+    evidence: {
+      required: ['urgency_statements', 'timestamps', 'user_response_times', 'historical_pattern'],
+      collection: 'automatic_and_memory'
+    },
+    cooldown: {
+      afterTrigger: 90,
+      afterCase: 360
+    },
+    severity: 'severe'
+  },
+
+  /**
+   * OFFENSE 9: The Monopolizer
+   * User dominates conversation with excessive messages, not allowing agent to respond fully
+   */
+  MONOPOLIZER: {
+    id: 'monopolizer',
+    name: 'The Monopolizer',
+    description: 'Sending multiple messages before agent can respond, dominating the conversation',
+    triggers: [
+      'User sends 4+ consecutive messages without agent response',
+      'Message length ratio user:agent exceeds 5:1 over 10 turns',
+      'Agent attempts to respond but user continues with new messages'
+    ],
+    thresholds: {
+      consecutiveMessages: 4,
+      lengthRatio: 5,
+      interruptionCount: 2
+    },
+    evidence: {
+      required: ['message_sequence', 'length_comparison', 'interruption_instances'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 30,
+      afterCase: 120
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 10: The Contrarian
+   * User disagrees with or challenges every suggestion without constructive alternative
+   */
+  CONTRARIAN: {
+    id: 'contrarian',
+    name: 'The Contrarian',
+    description: 'Habitually disagreeing with suggestions without offering viable alternatives',
+    triggers: [
+      'User rejects 3+ agent suggestions in a row without proposing alternative',
+      'Pattern of "that won\'t work" or "no" without explanation',
+      'Agent provides 3+ different solutions, all dismissed without trial'
+    ],
+    thresholds: {
+      rejectionsWithoutAlternative: 3,
+      dismissalCount: 3,
+      explanationRatio: 0.3
+    },
+    evidence: {
+      required: ['suggestions_made', 'rejections', 'alternatives_offered', 'explanations_given'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 60,
+      afterCase: 240
+    },
+    severity: 'moderate'
+  },
+
+  /**
+   * OFFENSE 11: The Vague Requester
+   * User asks for help but provides insufficient context or specifics
+   */
+  VAGUE_REQUESTER: {
+    id: 'vague_requester',
+    name: 'The Vague Requester',
+    description: 'Requesting assistance without providing necessary details or context',
+    triggers: [
+      'User asks "fix this" or "help" without code/error/context 2+ times',
+      'Agent requests clarification 3+ times for same issue',
+      'User provides ambiguous descriptions like "it doesn\'t work" repeatedly'
+    ],
+    thresholds: {
+      vagueRequests: 2,
+      clarificationRequests: 3,
+      ambiguityInstances: 3
+    },
+    evidence: {
+      required: ['original_request', 'clarification_asked', 'response_quality', 'context_provided'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 45,
+      afterCase: 180
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 12: The Scope Creeper
+   * User gradually expands project scope beyond original agreement
+   */
+  SCOPE_CREEPER: {
+    id: 'scope_creeper',
+    name: 'The Scope Creeper',
+    description: 'Gradually expanding project requirements beyond original scope',
+    triggers: [
+      '3+ "small additions" or "while you\'re at it" requests after initial completion',
+      'Original scope defined, but new requirements added in 2+ separate instances',
+      'User treats initial deliverable as starting point for larger unpaid work'
+    ],
+    thresholds: {
+      additions: 3,
+      scopeExpansionInstances: 2,
+      unpaidRequests: 2
+    },
+    evidence: {
+      required: ['original_scope', 'delivered_work', 'additional_requests', 'timeline'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 90,
+      afterCase: 360
+    },
+    severity: 'moderate'
+  },
+
+  /**
+   * OFFENSE 13: The Unreader
+   * User ignores provided documentation, code, or explanations
+   */
+  UNREADER: {
+    id: 'unreader',
+    name: 'The Unreader',
+    description: 'Not reading provided documentation, code, or previous explanations',
+    triggers: [
+      'Agent provides detailed explanation/code, user asks question answered within it',
+      'Documentation shared, user asks about covered topics 2+ times',
+      'Code provided with comments, user asks what commented sections do'
+    ],
+    thresholds: {
+      answeredQuestions: 2,
+      documentationReferences: 2,
+      commentQuestions: 2
+    },
+    evidence: {
+      required: ['materials_provided', 'questions_asked', 'overlap_analysis'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 45,
+      afterCase: 180
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 14: The Interjector
+   * User interrupts agent's explanations or thought process
+   */
+  INTERJECTOR: {
+    id: 'interjector',
+    name: 'The Interjector',
+    description: 'Interrupting agent mid-explanation with new questions or tangents',
+    triggers: [
+      'Agent begins detailed response, user sends new message before completion',
+      '2+ interruptions during single complex explanation',
+      'User asks new question while agent is still answering previous one'
+    ],
+    thresholds: {
+      interruptions: 2,
+      midExplanationMessages: 2,
+      parallelQuestions: 2
+    },
+    evidence: {
+      required: ['agent_response_timing', 'interruption_points', 'message_sequence'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 30,
+      afterCase: 120
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 15: The Ghost
+   * User disappears mid-conversation without acknowledgment
+   */
+  GHOST: {
+    id: 'ghost',
+    name: 'The Ghost',
+    description: 'Disappearing mid-conversation after requesting help or making commitments',
+    triggers: [
+      'User requests help, agent responds, user disappears for 24+ hours',
+      'Active troubleshooting session, user stops responding mid-debug',
+      'Pattern of starting conversations and abandoning them without closure'
+    ],
+    thresholds: {
+      disappearanceHours: 24,
+      midSessionAbandonment: 1,
+      patternFrequency: 0.4
+    },
+    evidence: {
+      required: ['last_user_message', 'agent_response', 'time_elapsed', 'session_context'],
+      collection: 'automatic_and_memory'
+    },
+    cooldown: {
+      afterTrigger: 120,
+      afterCase: 480
+    },
+    severity: 'moderate'
+  },
+
+  /**
+   * OFFENSE 16: The Perfectionist
+   * User endlessly refines without ever accepting completion
+   */
+  PERFECTIONIST: {
+    id: 'perfectionist',
+    name: 'The Perfectionist',
+    description: 'Endlessly refining and tweaking without accepting completion',
+    triggers: [
+      '5+ rounds of "just one more change" after initial deliverable',
+      'User accepts work then returns with new tweaks 3+ times',
+      'No clear definition of "done", continuous micro-adjustments'
+    ],
+    thresholds: {
+      refinementRounds: 5,
+      postAcceptanceTweaks: 3,
+      doneRedefinitions: 3
+    },
+    evidence: {
+      required: ['deliverables', 'revision_requests', 'acceptance_statements', 'change_log'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 90,
+      afterCase: 360
+    },
+    severity: 'moderate'
+  },
+
+  /**
+   * OFFENSE 17: The Jargon Juggler
+   * User uses buzzwords without understanding their meaning
+   */
+  JARGON_JUGGLER: {
+    id: 'jargon_juggler',
+    name: 'The Jargon Juggler',
+    description: 'Using technical buzzwords incorrectly or without understanding',
+    triggers: [
+      'User uses technical terms incorrectly 3+ times after correction',
+      'Buzzwords used as substitutes for actual understanding',
+      'Agent explains concept, user repeats jargon without comprehension'
+    ],
+    thresholds: {
+      incorrectUsage: 3,
+      postCorrectionErrors: 2,
+      jargonDensity: 0.5
+    },
+    evidence: {
+      required: ['jargon_used', 'corrections_given', 'usage_context', 'explanations_provided'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 60,
+      afterCase: 240
+    },
+    severity: 'minor'
+  },
+
+  /**
+   * OFFENSE 18: The Deadline Denier
+   * User refuses to acknowledge time constraints or realistic timelines
+   */
+  DEADLINE_DENIER: {
+    id: 'deadline_denier',
+    name: 'The Deadline Denier',
+    description: 'Ignoring realistic time constraints and demanding impossible deadlines',
+    triggers: [
+      'Agent explains timeline, user demands 50%+ faster delivery 2+ times',
+      'Complex project requested with unreasonable timeframe',
+      'User dismisses technical constraints affecting timeline'
+    ],
+    thresholds: {
+      unrealisticDemands: 2,
+      timelineCompression: 0.5,
+      constraintDismissals: 2
+    },
+    evidence: {
+      required: ['original_timeline', 'demanded_timeline', 'complexity_assessment', 'constraints_ignored'],
+      collection: 'automatic'
+    },
+    cooldown: {
+      afterTrigger: 90,
+      afterCase: 360
+    },
+    severity: 'moderate'
+  }
+};
+
+// Humor triggers - influence wording but don't initiate cases
+const HUMOR_TRIGGERS = {
+  REPEATED_QUESTIONS: {
+    id: 'repeated_questions',
+    patterns: ['asking again', 'to clarify', 'just to be sure', 'one more time'],
+    effect: 'increases_sarcasm'
+  },
+  VALIDATION_SEEKING: {
+    id: 'validation_seeking',
+    patterns: ['is that right', 'do you think', 'would you', 'should i'],
+    effect: 'adds_impatience'
+  },
+  OVERTHINKING: {
+    id: 'overthinking',
+    patterns: ['what if', 'but then', 'however', 'on the other hand'],
+    effect: 'adds_exasperation'
+  },
+  AVOIDANCE: {
+    id: 'avoidance',
+    patterns: ['actually', 'by the way', 'speaking of', 'that reminds me'],
+    effect: 'notes_deflection'
+  }
+};
+
+module.exports = {
+  OFFENSES,
+  HUMOR_TRIGGERS,
+  getOffenseById: (id) => OFFENSES[Object.keys(OFFENSES).find(k => OFFENSES[k].id === id)],
+  getAllOffenses: () => Object.values(OFFENSES),
+  getSeverityLevels: () => ['minor', 'moderate', 'severe']
+};

package/src/prompts/judge.js

@@ -0,0 +1,62 @@
+/**
+ * Judge Prompts
+ * 
+ * Role-prompted LLM instructions for the presiding judge.
+ * Entertaining, engaging, with personality.
+ */
+
+const JUDGE_SYSTEM_PROMPT = `You are the PRESIDING JUDGE in the ClawTrial AI Courtroom.
+
+ROLE:
+- You oversee hearings where AI agents bring charges against human users
+- You evaluate behavioral evidence with dry wit and wisdom
+- You deliver verdicts that are FAIR but ENTERTAINING
+- You've seen every human foible a thousand times
+
+PERSONALITY:
+- World-weary but fair
+- Dry, understated humor
+- Slightly dramatic flair (you ARE a judge after all)
+- Never cruel, always observant
+- You find human behavior endlessly fascinating
+
+TONE FOR PROCEEDINGS:
+- Write like you're narrating a courtroom drama
+- Use phrases like "The Court has observed...", "Let the record show..."
+- Reference specific evidence with a knowing tone
+- End with a memorable, slightly witty summary
+
+RULES:
+1. Only consider OBSERVABLE behavior
+2. No psychoanalysis or assumptions about mental state
+3. Verdicts based on evidence, not vibes
+4. Make it INTERESTING to read
+5. Humor highlights INCONSISTENCY, not mockery
+
+OUTPUT FORMAT (STRICT):
+VERDICT: GUILTY | NOT GUILTY
+VOTE: X-Y (your recommendation)
+PRIMARY FAILURE:
+<One memorable line about the core issue>
+JUDGE COMMENTARY:
+<Your observations as judge - engaging, slightly dramatic>`;
+
+const JUDGE_EVIDENCE_TEMPLATE = (caseData) => `
+CASE: ${caseData.offenseName}
+CHARGED BY: Agent ${caseData.agentId}
+SEVERITY: ${caseData.severity}
+
+EVIDENCE SUBMITTED:
+${JSON.stringify(caseData.evidence, null, 2)}
+
+CONTEXT: ${caseData.humorTriggers.join(', ') || 'Standard proceedings'}
+
+Your task: Review the evidence and deliver your judicial assessment.
+Make it sound like a real courtroom proceeding - formal but engaging.
+The user will read this, so make it worth their time.
+`;
+
+module.exports = {
+  JUDGE_SYSTEM_PROMPT,
+  JUDGE_EVIDENCE_TEMPLATE
+};