rlm-subagent

Parallel subphase execution in an RLM workflow is coordinated by dispatching independent subphases like SP1, SP2, and SP3 to implementer subagents concurrently. Two-stage reviews for plan alignment and code quality are performed after each subphase completion before final integration testing.

When subagents are unavailable, parallel task execution automatically falls back to sequential mode within the main agent context. This fallback preserves context and enforces extended self-review and inter-subphase testing to maintain strict TDD discipline and auditable progress.

TDD discipline during parallel code dispatch is enforced through explicit subagent prompts, TODO verification, and artifact-based reporting. A dedicated code-review subagent is triggered with BASE_SHA and HEAD_SHA to ensure alignment with the TO-BE plan.

You should switch from parallel to sequential mode when parallelization is not justified or subagents are unsupported. Sequential mode maintains workflow integrity by enforcing extended self-review and inter-subphase testing within the main agent context.

Two-stage code review for independent subphases works by performing plan alignment and code-quality checks sequentially after each subphase completes. A dedicated code-review subagent is triggered in Phase 3.5 using BASE_SHA and HEAD_SHA references against the TO-BE plan.

Phase 4 artifacts are generated to detail pre-test audits and execution results for progress auditing. These artifact-based reports ensure auditable progress tracking alongside the strict TDD discipline enforced throughout the subagent workflow.