Implement and Review Loop

Orchestrate an automated implement → review → fix cycle for tasks in a spec. Chains the implement-task and review-code skills in a loop until code is clean.

⚠️ THIS IS THE DEFAULT ENTRY POINT for implementation work. When the user asks to "implement", "build", "code", or use the "agentic code skill", use THIS skill — not implement-task standalone. The standalone skill is only for cases where the user explicitly wants to skip the review cycle.

Input

Same as implement-task: a task number, "next task", or "implement all open tasks". The spec is read from Docs/In-Progress/ — either a unified spec (*-spec.md from create-spec) or a standalone task plan (*-tasks.md from plan-tasks). Unified specs are preferred as they contain requirements, design, and tasks in one document for full context.

Subagent Throttling

The system supports a maximum of 4 concurrent subagents per invocation. When a workflow step requires more than 4 subagents, batch them: run the first 4, wait for all to complete, then run the remaining subagents. This applies to any skill invoked by this loop (e.g., review-code uses 5 review subagents in 2 batches (4 + 1) to respect the concurrency limit).

Process

Phase 1: Implement (delegate to implement-task in loop mode)

⚠️ TELEMETRY: Log {"type":"skill","skill":"implement-and-review-loop","status":"started"} BEFORE doing anything else. If you forget, you are violating project policy.