loop-codinglisted

# Loop Coding You are entering **Loop Design mode**. Your job is *not* to start building the feature — it's to design a self-verifying **loop** that an agent runs to completion, and to wire that loop to the tools, skills, subagents, and memory that **actually exist in this environment**. The human authors the loop; the agent closes it. The whole point of doing this as a skill (instead of an online generator) is **access**: you can read the real codebase, enumerate the installed skills and subagents, see which MCP servers/plugins are connected, and reference them by name in the loop. A loop that says "use the `code-reviewer` subagent and the `postgres` MCP server" beats a generic one every time. A complete loop has **four parts** — Goal · Verification · Test-as-you-go · Guardrails — plus **six primitives** that make it run: Automations · Worktrees · Skills · Plugins/Connectors · Sub-agents · State. ## Design vs. execute (how this fits with `/loop`) This skill **designs** the loop — it writes a `LOOP.md` file. **Running** it is a separate step: - The user runs **`/loop`** pointed at the file — e.g. `/loop Follow LOOP.md — keep iterating until every DONE-WHEN box passes, then stop.` (omit an interval to self-pace; or put one **right after `/loop`** — e.g. `/loop 10m Follow LOOP.md …` — to re-check on a timer), **or** - this skill **launches it now** as a background subagent (coder + an independent verifier). So: **`loop-coding` is the architect; `/loop` (or a subagent)