review-queuelisted

# Review Queue — isolated per-PR reviews, fanned out and synthesized The problem this solves: reviewing N peer PRs in one long session pollutes context — findings, file reads, and design intent from PR A bleed into PR B. This skill keeps each review in its **own sub-agent + own git worktree**, then consolidates. One command, N clean reviews, one report. **Model:** the orchestrator (you, the main session) builds the worklist, dispatches one `pr-reviewer` sub-agent per PR, and synthesizes. The sub-agent carries the discipline (pr-review-kit) and `isolation: worktree`; the orchestrator never reviews a PR itself (that would re-pollute the main context). ## Arguments A list of PR numbers/URLs to review (e.g. `42088 42090 39311`, or full `chainguard-dev/mono#42090` refs). If empty, the skill builds the queue from your review inbox (Phase 0). --- ## Phase 0 — Build the worklist + catch up 1. **Get the queue** (first source that's available): - **zen** (if the `zen` MCP is connected): call `zen_inbox` — PRs awaiting *you* (configured authors + explicit review requests). This is the intended source. - **gh** fallback: `gh search prs --review-requested=@me --state=open --json number,title,repository` (and/or the repos you care about). - **args**: if the user passed PR refs, use those verbatim — skip discovery. 2. **Catch up via an observational memory layer** (optional — only if your harness provides one, e.g. a memory/timeline MCP) — for each PR you're about to review,