/team-implement — Parallel Implementation with Adversarial Review

Spawn implementation subagents working on different crates in parallel, with separate review subagents.

Execution

1. •Read the approved plan from working/plans/
2. •Partition work by crate — each subagent owns specific files
3. •Spawn implementation Task calls in the same response:

Per crate/module:

Task(subagent_type="production-code-engineer",
     prompt="You own crates/{crate}/. Implement the following from the plan:\n{task_description}\n\nRules:\n- Only modify files in your assigned crate\n- Follow TDD: write failing tests first, then implement\n- Run: cargo build -p {crate} && cargo test -p {crate}\n- Report what you changed and test results")

1. •After all implementers complete, spawn review subagent:

Task(subagent_type="senior-code-reviewer",
     prompt="<content of .claude/agents/code-reviewer.md>\n\nReview the changes in:\n{changed_files}")

1. •If reviewer finds Critical/Major issues, route fixes to the responsible implementer (max 3 rounds)
2. •Run make dev on combined result
3. •Present summary

Arguments

• •/team-implement — use most recent approved plan
• •/team-implement working/plans/2026-02-08_feature.md — specific plan

File Ownership Rules

• •Each subagent owns specific crate(s) — no overlapping file edits
• •Root config files (Cargo.toml, Makefile): only main context modifies
• •Iron Rule: The subagent that writes code NEVER reviews it