feat: Wave Planner - file-aware parallel agent orchestration

[mmorris35]

Problem

When DevPlan generates tasks executed by parallel agents within a wave, agents may modify the same files simultaneously, causing merge conflicts. Currently wave planning is ad-hoc with no file dependency analysis.

Observed in: g-hive Wave 3.4.2 - parallel agents created merge conflicts on shared files.

Proposed Solution

Add a plan_waves MCP tool that:

1. Analyzes task-to-file dependencies - infer which files each task will touch
2. Assigns file ownership - prevent parallel contention
3. Generates orchestrated wave plans - explicit constraints for agent spawners

New Tool: plan_waves

Input:

• Task list with descriptions
• Project file tree
• Max parallel agents
• Conflict strategy (sequence/partition/delegate)

Output:

• Wave assignments with file ownership per agent
• Dependency ordering for shared files
• Conflict risk assessment

Conflict Strategies

• sequence: Tasks touching shared files run sequentially (safe, slower)
• partition: Split shared files if semantically possible
• delegate: Flag conflicts for human/integration agent resolution

Implementation Phases

1. Static Analysis - Pattern matching, AST analysis for file dependencies
2. Opus-Assisted - LLM inference for ambiguous tasks
3. Orchestration Output - Structured JSON for agent spawners

Success Criteria

• Zero merge conflicts when following wave plan
• Maximum parallelization where file deps allow
• Clear audit trail of file ownership

Spec

Full specification: docs/wave-planner-spec.md

---

Lesson learned from g-hive development - parallel agents within same wave need file-aware orchestration.