agentsclimarketplace

Review inbox

Skill HsuanYuLee/polaris/.claude/skills/review-inbox

Use when the user wants to discover and review PRs across the team awaiting their attention. NOT for a single specific PR (use review-pr). Supports three discovery modes: Label (GitHub label scan), Slack (channel-wide scan), Thread (specific Slack thread URL). Trigger: '掃 PR', 'review 大家的 PR', '批次 review', '有哪些 PR 要我看', Slack thread URL + review intent ('review <slack_url>', '幫我看這串'). Key: '我的 PR' → check-pr-approvals; '大家的 PR' / Slack URL → here; single PR URL → review-pr.From its SKILL.md

Install
npx -y skills add HsuanYuLee/polaris --skill review-inbox

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 6 commands, including `annotate-review-candidates.py` and 5 more.

SKILL.md

5.9 KB, ~1.5k tokens by cl100k_base, as published. Nobody here has run it

Review Inbox — 批次 Review 待審 PR

找出 team 內需要自己 first review、re-approve、或 re-review 的 PR,批次執行 review, 並依來源回 Slack 通知。

Contract

此 skill 只處理多 PR discovery + batch review orchestration。單一 PR URL 轉 review-pr; 「我的 PR」approval 狀態轉 check-pr-approvals。

支援三個來源:

SourceUse when
Slack預設;掃 PR channel 最近訊息中的 PR URLs
Thread使用者提供 Slack thread URL 並要求 review
Label使用者明確提到 need review label

不得 review 自己的 PR。不得對 waiting_for_author PR 重複 review。 Review inbox 屬 reviewer-side read-only lane;它可以 advisory,但對 awaiting_re_review、 mergeable_ready、unsupported_mutation、changes_requested 的解釋必須沿用 shared PR state。 它不得把 batch review 結果升格成 author-side completion / release authority;所有 「可 merge / 已修完 / 可 release」語句只能轉述 shared state,不可自行推論。

Reference Loading

SituationLoad
Any runcontext-budget-contract.md, review-inbox-discovery-flow.md, shared-defaults.md, stale-approval-detection.md, workspace-config-reader.md
Batch review executionreview-inbox-batch-review-flow.md, review-inbox/dispatch-context-bundle.md
Slack notificationreview-inbox-slack-reporting.md, slack-message-format.md, github-slack-user-mapping.md, workspace-language-policy.md

Slack channel scan 可以派 read-only sub-agent。Per-PR review 不得使用 Claude Code general-purpose sub-agent;只有 runtime 提供 constrained code-reviewer adapter 時才可平行 dispatch。否則依 build-review-runtime-plan.py 產生的 main_session_sequential plan,一次執行 一個 review packet,完成後只把 Completion Envelope summary 留在主 context。Batch review dispatch 由 main session 讀取 dispatch-context-bundle.md 一次,再把濃縮後的 review flow inline 注入每個 review packet;不得要求執行者重讀完整 review skill / reference stack。

Flow

  1. 讀 workspace config 與 defaults,取得 GitHub org、PR channel、approval threshold、 batch size、concurrency、confirm setting。
  2. 解析 mode:Thread 優先,其次 explicit Label,其餘走 Slack。
  3. 取得 current GitHub username,作為 exclude author 與 review-status 判定依據。
  4. 依 discovery reference 產生 candidates JSON;scan snapshot 超過 60 秒不可沿用。 Slack channel scan 使用 MCP 時指定 concise output;fallback CLI 的 --oldest 可接受 Slack timestamp 或 ISO date/datetime。
  5. 將 candidates JSON 經 annotate-review-candidates.py enrich,補上 sister PR cluster metadata 與 model_tier semantic class。Slack mapping 若含 root_ticket_key,cluster 必須優先使用 root ticket;若沒有 umbrella ticket 但同一 Slack root message 有可辨識 topic,使用 root_topic_key;最後才 fallback 到每張 PR 自己的 ticket。
  6. 若 candidates 為空,回報目前沒有需要 review 的 PR 並停止。
  7. 顯示排序後清單;若 config 要求 confirm,等待使用者選擇。
  8. 先用 build-review-prompt.sh 產生 review packets + manifest,再用 build-review-runtime-plan.py 產生 runtime plan。Plan 必須禁止 general-purpose sub-agent; 若無 constrained code-reviewer adapter,主 session 依 sequential plan 執行。
  9. 依 batch size / runtime plan 執行 per-PR review packets;prompt 必須使用 deterministic handbook resolver 列出已存在的 project handbook paths,空清單時明確標示 no project handbook。Prompt 必須要求執行者先讀 changed-file names,再依 diff size sampling;existing inline comments 只能以 metadata-only dedup,不把完整 comment body 放進 context。Cluster lead 先跑完整 review;cluster sibling 使用 sibling-diff mode 與 small_fast model class hint,不確定時標記 needs_standard_review。
  10. 收斂結果,依來源模式發 Slack summary 或 thread replies。
  11. 跑 measure-review-inbox-session.sh 產生 telemetry JSON,並用 polaris-learnings.sh add --type telemetry --tag review-inbox 寫入 metadata.review_inbox_run。若無法取得完整 transcript,仍需用 line-count proxy 記錄 candidate count、reviewed count、runtime plan kind 與 artifact volume。
  12. 在對話中回報每個 PR 的 review result、approve status 與 telemetry run_id。

Write And Notification Rules

  • Review body 與 inline comments 由 review-pr 流程負責。
  • Slack message 送出前必須通過 workspace-language-policy.md language gate。
  • Slack mode 不發 channel-wide summary;只回覆原始 PR threads。
  • Label mode 發一則 channel summary。
  • Thread mode 回覆指定 thread。
  • Thread replies 不設 reply_broadcast: true。

Completion

輸出 reviewed count、APPROVE / REQUEST_CHANGES / COMMENT counts、Slack notification count、 任何 blocked PR 的原因,以及 telemetry run_id。Telemetry 可用:

POLARIS_WORKSPACE_ROOT={workspace_root} .claude/skills/references/scripts/polaris-learnings.sh query --type telemetry --tag review-inbox

查到 metadata.review_inbox_run.main_session_input_tokens。

Post-Task Reflection (required)

Execute post-task-reflection-checkpoint.md before reporting completion.

What ships with it: 40 files

220.9 KB alongside SKILL.md, 29 of them executable

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.