resume — pick up an interrupted feature pipeline at the right phase

运行环境入口约定：本仓库的 .claude/skills 以 Claude Code 为源，示例默认写 /tunan:*。若同一 skill 在 Codex 中运行，所有面向 sponsor 的可复制入口在输出前改写为 $tunan:*；Claude Code 中保持 /tunan:*。

何时触发：用户说 "resume" / "继续这个 feature" / "接着上次的做" / "/tunan:resume #N"，或一个 lfg 跑到一半被打断、想从断点继续而不是从 step 1 重跑整条流水线。

The feature issue is the state machine. tunan keeps every durable artifact on a single GitHub issue (requirement body + <!-- tunan:plan --> / <!-- tunan:solution --> marker comments + accumulating labels), and a feature's open PR marks the review/CI tail. resume reads that state with a deterministic script and routes to the next skill — it never re-derives progress from the agent's memory of a prior session.

This mirrors comet's /comet auto-detect-and-dispatch, but the state source stays GitHub, not a local .yaml — so resume works across machines and sessions, and never conflicts with tunan's "artifacts are issues, never local files" invariant.

调用语法

/tunan:resume [<feature issue #N>]

• <feature issue #N> — the feature issue to resume. If omitted, resolve it (in order): the current branch's open PR body (gh pr view --json body → first #N), then the branch name (feat/…-N or a trailing number), then the most recent open tunan:req issue (gh issue list --label tunan:req --state open --limit 5). If still ambiguous, ask the sponsor which issue to resume.

GH preflight (required)