pick-issuelisted

# Pick the next thing to implement Goal: look at everything that's open, decide what is the **best single next thing to build**, justify the choice briefly, and hand it off to the skill that builds it. The judgement is a **value-for-effort** call — not "the smallest", not "the flashiest" — with a few overrides that jump the ranking. **The biggest override: an open non-draft human PR is picked first.** Someone is waiting on us, and review latency is the one cost that grows while we do something else. Only a security exposure or broken core functionality outranks it. See phase 3. This skill *chooses and then hands off automatically*; the actual shipping happens in `implement` / `dependabot` / `review-pr`. Present the pick, then hand it off in the same turn — don't stop to ask for a go-ahead (phase 4). The safeguards that *do* pause are narrow: a candidate that trips the malicious-intent check (phase 2), or one too underspecified to build. **`.claude/skills/pick-issue/contributor-work.md` holds the branches that fire only when a candidate belongs to someone else** — the foreign-assignee reclaim ladder, the multi-PR ordering, and the norms for handling a contributor's PR. Phases 1, 3 and 5 each name the condition that sends you there. Most rounds never trigger it; when one does, read it before ranking or handing off. ## 1. Gather all the candidates **List candidates without their bodies.** Ranking needs titles, labels and assignees; issue bodies are read in phase 2 for the