cover-melisted

# Cover Me ## What this does Spawns the **`supervisor`** subagent in a **fresh, forked context** to review the work at a checkpoint you choose, the way you'd call a senior peer over at a risky moment. It re-derives the task cold, unanchored from your reasoning, and tells you whether you're on track or about to step on a rake. It is **non-blocking and advisory**: it flags, you decide. It never edits, never takes the wheel. It is a *cold review*, not a live watch. A subagent gets one fresh context, reads what you hand it, reports once, and terminates; there is no channel from it into a running agent. That is why the brief carries the distilled session transcript - evidence the reviewer did not get from you. Its edge is *isolation*, not magic: the builder is anchored to its own reasoning, so a same-context self-review just confirms. A fresh context can disagree, which is exactly how you catch drift and context rot. **The user decides when it runs.** Outside an explicit `/cover-me` or an equivalent request, never spawn the supervisor unprompted: when the situation calls for it - risky change, long context, visible frustration, a hook nudge - say so and ask in one line, then spawn only on a yes. A review the user did not know about is a violation of this skill, not a use of it. ## When to reach for it - **Mission-critical or irreversible work**: migrations, deletes, deploys, schema or auth changes, anything hard to undo. - **Before declaring "done"**: a pre-flight check t