awb-handoverlisted

# Package work for a fresh session > **Announce on activation:** "Using awb-handover — I'll package this into artifacts a cold reader can execute." The failure this prevents: a session ends with everything still in your head, and the next session — a teammate, or you in a week — re-derives the same decisions, contradicts them, or silently drops them. A handover is only as good as what a reader with *zero context* can do with it. This skill **packages** what's settled; it does not re-decide it. ## Scope - **Does:** crystallize already-made decisions and current state into artifacts a fresh agent can act on, and prove they're self-contained with a cold-reader test. - **Does NOT:** re-open settled decisions (only if the cold-reader test exposes a real gap), do new research (use a research skill first), or start implementing (hand off to a plan/implement skill). ## The four artifacts Split by *what each one answers*, so the next session reads only what it needs: | Artifact | Answers | Holds | |---|---|---| | `HANDOVER.md` | "Where are we?" | State: what's done, what's in progress, key decisions **with their rationale**, and dead ends not to retry | | `plan.md` | "What do we do?" | The ordered steps to execute, naming the files each one touches | | `spec.md` | "What does done look like?" | Acceptance criteria — the observable contract, not the implementation | | `NEXT_SESSION_PROMPT.md` | "How do we start?" | The literal paste-in prompt: which files to read first, the