feat: add parallel execution extension

[aymanyaq]

Summary

Added new optional extension to the AI-DLC rules that introduces adaptive, accuracy-first parallel execution for both greenfield and brownfield multi-unit projects.

For full context and discussion, please see the RFC: #218

Changes

• Parallel Execution (19 rules, PARALLEL-EXEC-001 through PARALLEL-EXEC-019): A comprehensive, coordinator-driven parallel execution model that speeds up multi-unit projects while maintaining accuracy as a non-negotiable constraint.

Key Capabilities:

• Wave Dependency Planning (001): Analyzes unit dependencies, groups independent units into waves, and performs accuracy safety assessments with explicit safety matrices before any parallel work begins
• Pre-Flight Accuracy Gate (002): Re-validates wave safety at execution time — catches conditions that changed since the plan was written
• Sub-Task Parallelism with Convergence (003): Enables within-unit parallelism (e.g., backend ∥ frontend) with a mandatory convergence gate checking API contract and test alignment
• Anticipatory Test Planning (004): Begins test plan writing immediately after Functional Design approval (safe because tests derive from design, not code)
• Cross-Unit Knowledge Sharing (005): Propagates confirmed patterns via shared-patterns.md so later units don't re-solve problems or introduce inconsistencies
• Coordinator Dispatch Model (006): Defines the full parallel dispatch lifecycle — prompt templates, context discipline, result collection, unresolved question consolidation, and fallback-to-sequential
• Single-Unit Dormancy (007): Gracefully enters dormant mode for single-unit projects
• Scope Expansion Trigger (008): Structured path for splitting units mid-construction when complexity warrants it
• Alternative Sub-Task Archetypes (009): Extends sub-task parallelism beyond backend/frontend to data-pipeline ∥ visualization, ML training ∥ serving, independent feature modules, and IaC ∥ application code
• Data Contract in Dispatch (010): Ensures subagents receive data profile paths for data-driven apps
• Shared Dependency Audit (011): Ensures brownfield dependency context reaches subagents
• Runtime Convergence Validation (012): Catches hardcoded values that don't match actual data — the most dangerous class of parallel execution failure
• Wave Grouping / Critical Path / Priority (013-015): Optimizes wave composition and prioritizes critical-path units
• Batched Functional Design (016): Produces FD for all wave units together for cross-unit consistency
• Wave Integration Checkpoint (017): Mandatory integration verification between waves
• Global NFR with Parallel Waves (018): Ensures NFR coverage for units that skip dedicated NFR stages
• Session Resume (019): Handles session continuity with context loading discipline for parallel dispatch

Core Design Principles:

• Accuracy First: Default is sequential. Parallel must be earned through formal safety assessment
• Three accuracy threats that must never occur: Divergent assumptions, stale context, premature work
• Adaptive: Falls back to sequential at any point if accuracy risk is detected
• Platform-agnostic: Uses generic parallel dispatch terminology, not platform-specific APIs

User experience

This extension is opt-in. When a user is in the Requirements Analysis phase, they will be prompted with opt-in questions to enable parallel execution.

Checklist

• I have reviewed the contributing guidelines
• I have performed a self-review of this change
• Changes have been tested
• Changes are documented

Test Plan

• Load the extension during the Requirements Analysis phase and verify the opt-in prompt appears.
• All markdown files have passed markdownlint-cli2.

Acknowledgment

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the project license.