UNPKG

aiwg

Version:

Deployment tool and support utility for AI context. Copies agents, skills, commands, rules, and behaviors into the paths each AI platform reads (Claude Code, Codex, Copilot, Cursor, Warp, OpenClaw, and 6 more) so one source of truth works across 10 platfo

303 lines (242 loc) 6.37 kB
# Criticality-Based Panel Sizing Rules **Enforcement Level**: MEDIUM **Scope**: Ensemble review pattern selection **Research Basis**: REF-017 Self-Consistency **Issue**: #161 ## Overview These rules define how task criticality determines ensemble review panel sizes, ensuring high-stakes decisions receive proportionally thorough review while routine changes remain efficient. ## Research Foundation From REF-017 Self-Consistency (Wang et al., 2022): - Panel size impacts both accuracy and cost - 5 reviewers provide optimal accuracy/cost balance for most tasks - Critical decisions benefit from larger panels - Low-risk changes can use smaller panels efficiently ## Criticality Levels ### Level Definitions | Level | Panel Size | Threshold | Description | |-------|-----------|-----------|-------------| | **CRITICAL** | 7 | 0.85 | Irreversible changes, security, compliance | | **HIGH** | 5 | 0.80 | Architecture, public APIs, data models | | **STANDARD** | 5 | 0.60 | Feature implementation, refactoring | | **LOW** | 3 | 0.67 | Documentation, minor fixes, config | ### Automatic Criticality Detection The following patterns trigger automatic criticality escalation: #### CRITICAL Triggers ```yaml critical_triggers: file_patterns: - "**/security/**" - "**/auth/**" - "**/encryption/**" - "**/compliance/**" - "**/*.pem" - "**/*.key" - "**/secrets/**" content_patterns: - "DELETE.*CASCADE" - "DROP TABLE" - "TRUNCATE" - "rm -rf" - "force push" - "breaking change" artifact_types: - security_gate - compliance_attestation - data_deletion_request ``` #### HIGH Triggers ```yaml high_triggers: file_patterns: - "**/api/**" - "**/schema/**" - "**/migration/**" - "**/architecture/**" - "**/*.proto" - "**/contracts/**" artifact_types: - architecture_decision_record - api_contract - database_migration - public_interface ``` #### LOW Triggers ```yaml low_triggers: file_patterns: - "**/docs/**" - "**/*.md" - "**/README*" - "**/.gitignore" - "**/comments/**" artifact_types: - documentation_update - comment_addition - typo_fix - formatting_change ``` ## Panel Configuration by Criticality ### CRITICAL Configuration ```yaml critical_review: panel_size: 7 threshold: 0.85 timeout_minutes: 60 reviewer_composition: required: - security_auditor - architecture_designer - domain_expert optional: - test_engineer - code_reviewer - compliance_officer - technical_writer confidence_levels: high: threshold: 0.90 action: accept medium: threshold: 0.80 action: flag_for_human_review low: threshold: 0.70 action: require_human_decision escalate: threshold: 0.70 action: escalate_to_stakeholders escalation: on_low_confidence: true on_any_dissent: true human_required: true ``` ### HIGH Configuration ```yaml high_review: panel_size: 5 threshold: 0.80 timeout_minutes: 30 reviewer_composition: required: - architecture_designer - code_reviewer optional: - test_engineer - domain_expert - technical_writer confidence_levels: high: threshold: 0.85 action: accept medium: threshold: 0.70 action: accept_with_note low: threshold: 0.60 action: flag_for_review escalate: threshold: 0.60 action: escalate_to_critical escalation: on_low_confidence: true expand_to_panel: 7 ``` ### STANDARD Configuration ```yaml standard_review: panel_size: 5 threshold: 0.60 timeout_minutes: 15 reviewer_composition: required: - code_reviewer optional: - test_engineer - architecture_designer - technical_writer - domain_expert confidence_levels: high: threshold: 0.80 action: accept medium: threshold: 0.60 action: accept_with_note low: threshold: 0.50 action: flag_for_review escalate: threshold: 0.50 action: expand_panel ``` ### LOW Configuration ```yaml low_review: panel_size: 3 threshold: 0.67 timeout_minutes: 5 reviewer_composition: required: - code_reviewer optional: - technical_writer - test_engineer confidence_levels: high: threshold: 0.80 action: accept medium: threshold: 0.67 action: accept low: threshold: 0.50 action: escalate_to_standard ``` ## Criticality Override ### Manual Override Humans can override automatic criticality: ```markdown --- review: criticality: CRITICAL override_reason: "Contains sensitive customer data logic" --- ``` ### Override Audit All criticality overrides MUST be logged: ```yaml override_log: artifact: "path/to/file.ts" detected_criticality: STANDARD override_criticality: CRITICAL reason: "Contains sensitive customer data logic" override_by: "human_reviewer" timestamp: "2026-01-25T10:00:00Z" ``` ## Integration with Ensemble Review This criticality system integrates with `@$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/schemas/flows/ensemble-review.yaml`: 1. **Pattern Selection**: Criticality determines which review pattern to use 2. **Panel Sizing**: Panel size scales with criticality 3. **Threshold Adjustment**: Higher criticality = stricter thresholds 4. **Escalation Path**: Low confidence at any level can escalate upward ## Validation ### Pre-Review Validation Before initiating ensemble review: - [ ] Criticality level determined (auto or manual) - [ ] Panel size matches criticality - [ ] Required reviewers available - [ ] Timeout configured appropriately ### Post-Review Validation After completing ensemble review: - [ ] Agreement meets threshold for criticality - [ ] Escalation rules applied if needed - [ ] Override reasons documented if applicable - [ ] Audit trail complete ## References - @$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/schemas/flows/ensemble-review.yaml - Ensemble review patterns - @.aiwg/research/findings/REF-017-self-consistency.md - Research foundation - @.aiwg/research/synthesis/topic-05-verification.md - Verification patterns - #161 - Implementation issue --- **Rule Status**: ACTIVE **Last Updated**: 2026-01-25