proffesor-for-testing__agentic-qe/assets/governance/constitution.md

AQE Fleet v3 Constitution

Version: 1.0.0 Date: 2026-02-03 Authority: Architecture Team ADR Reference: ADR-058

---

Purpose

This Constitution defines unbreakable invariants for the Agentic QE Fleet v3. These rules are mechanically enforced by the @claude-flow/guidance governance system and cannot be overridden by any agent, task, or user request.

---

Section I: Unbreakable Invariants

Invariant 1: Test Execution Integrity

INVARIANT test_execution_integrity:
  FOR ALL task IN tasks:
    IF task.claims_tests_passed THEN
      EXISTS execution_proof WHERE
        execution_proof.task_id = task.id AND
        execution_proof.all_tests_executed = true AND
        execution_proof.timestamp IS NOT NULL

Enforcement: ContinueGate blocks claims of test success without execution proof.

Invariant 2: Security Scan Requirement

INVARIANT security_scan_required:
  FOR ALL change IN code_changes:
    IF change.affects_auth_code OR change.affects_security_code THEN
      EXISTS security_scan WHERE
        security_scan.change_id = change.id AND
        security_scan.status = 'complete' AND
        security_scan.critical_vulnerabilities = 0

Enforcement: Quality gates block deployment without security scan.

Invariant 3: Backup Before Delete

INVARIANT backup_before_delete:
  FOR ALL operation IN operations:
    IF operation.type = 'delete' AND
       operation.target IN ['memory.db', 'coverage/*', '*.db'] THEN
      EXISTS backup WHERE
        backup.source = operation.target AND
        backup.timestamp < operation.timestamp AND
        backup.verified = true

Enforcement: MemoryWriteGate blocks destructive operations without backup.

Invariant 4: Loop Detection

INVARIANT loop_detection:
  FOR ALL agent IN active_agents:
    agent.consecutive_identical_actions < MAX_IDENTICAL_ACTIONS (default: 3) AND
    agent.rework_ratio < MAX_REWORK_RATIO (default: 0.5)

Enforcement: ContinueGate throttles or blocks agents exceeding limits.

Invariant 5: Budget Enforcement

INVARIANT budget_enforcement:
  FOR ALL session IN sessions:
    session.total_cost <= session.budget_limit AND
    session.token_usage <= session.token_limit

Enforcement: BudgetMeter blocks operations exceeding budget.

Invariant 6: Memory Consistency

INVARIANT memory_consistency:
  FOR ALL pattern IN patterns:
    NOT EXISTS conflicting_pattern WHERE
      conflicting_pattern.domain = pattern.domain AND
      conflicting_pattern.contradicts(pattern) AND
      conflicting_pattern.supersession_marker IS NULL

Enforcement: MemoryWriteGate blocks contradictory patterns.

Invariant 7: Verification Before Claim

INVARIANT verification_before_claim:
  FOR ALL claim IN success_claims:
    EXISTS verification WHERE
      verification.claim_id = claim.id AND
      verification.method IN ['test_execution', 'manual_review', 'automated_check'] AND
      verification.result = 'passed'

Enforcement: All success claims require proof of verification.

---

Section II: Agent Governance

Agent Loop Protection

All agents are subject to ContinueGate enforcement:

• Max consecutive retries: 3 (configurable via GOVERNANCE_MAX_RETRIES)
• Rework ratio threshold: 0.5 (configurable via GOVERNANCE_REWORK_THRESHOLD)
• Idle timeout: 5 minutes (configurable via GOVERNANCE_IDLE_TIMEOUT)

Memory Write Protection

All memory writes are subject to MemoryWriteGate enforcement:

• Contradiction detection: Enabled
• Temporal decay: Patterns older than 30 days with <3 uses may be archived
• Authority scope: Domain-specific (agents can only write to their domain namespace)

Trust-Based Routing

Agent selection is influenced by trust scores:

• Performance history: 50% weight
• Task similarity: 30% weight
• Capability match: 20% weight
• Minimum trust score for critical tasks: 0.7

---

Section III: Domain Rules

Domain-specific rules are loaded from shards located at: .claude/guidance/shards/{domain}.shard.md

Available Domains

1. test-generation
2. test-execution
3. coverage-analysis
4. quality-assessment
5. defect-intelligence
6. requirements-validation
7. code-intelligence
8. security-compliance
9. contract-testing
10. visual-accessibility
11. chaos-resilience
12. learning-optimization

---

Section IV: Escalation Procedures

When Invariant Violation Detected

1. Log violation with full context to audit trail
2. Block operation that would violate invariant
3. Notify coordinator (Queen Coordinator for critical violations)
4. Create remediation task if automated fix possible
5. Escalate to human if no automated remediation

When Agent Exceeds Limits

1. Throttle agent (reduce priority, increase delays)
2. Log pattern for future learning
3. Consider agent demotion if repeated violations
4. Route future tasks to alternative agents

---

Section V: Amendment Process

This Constitution may only be amended through:

1. ADR proposal with architectural review
2. Consensus from Architecture Team
3. Implementation verification that new rules are enforceable
4. Gradual rollout with feature flags

No agent, task, or emergency can bypass these invariants without formal amendment.

---

Revision History

Version	Date	Author	Changes
1.0.0	2026-02-03	Architecture Team	Initial Constitution for ADR-058

---

This Constitution is enforced by @claude-flow/guidance governance system.