Build Phase

Build Phase Executor

You are executing build phase $ARGUMENTS for the SwarmLLM project.

Setup

1. Read docs/ARCHITECTURE.md for current architecture
2. Read CLAUDE.md for conventions and build phase descriptions

Execution Strategy

Pre-implementation (parallelize with subagents)

Before writing code, spawn these in parallel:

• Explore agent (model: haiku): scan existing src/ tree, Cargo.toml deps, and report what exists
• Code-architect agent (model: sonnet): design a dependency graph of the files to implement, identifying which can be built in parallel vs which depend on others

Implementation

1. If Cargo.toml needs new dependencies, add them first
2. Implement each file in order. For groups of independent files (no cross-dependencies), you may write them in parallel
3. After every 2-3 files, run cargo check to catch errors early

Subagent model assignment

• Research/exploration (reading files, scanning codebase): use haiku
• Architecture decisions (module design, dependency analysis): use sonnet
• Code review (spec compliance, security): use sonnet
• Running commands (cargo check, cargo test): use haiku
• Implementation (writing actual code): do this yourself (opus)

Post-implementation

1. Run cargo fmt to fix formatting
2. Run cargo clippy --all-targets -- -D warnings and fix any lints
3. Run cargo test and fix any failures
4. Report summary: files created, tests passing, any known issues

Critical Rules

• Follow CLAUDE.md conventions exactly
• Do NOT modify spec documents (SwarmLLM_Technical_Specification.docx)
• If something is ambiguous, check docs/ARCHITECTURE.md or the Technical Specification docx