Token导航 LogoToken导航TokenDH.com
研究检索需要联网github未标认证来源可访问许可证需确认审计提醒

relay-merge中继合并

Agent Skill

relay-merge 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

303

周安装

13

GitHub Stars

公开资料未说明

下载量

106
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:relay-merge(中继合并)
来源仓库:https://github.com/sungjunlee/dev-relay
仓库路径:skills/relay-merge
安装命令:
npx skills add https://github.com/sungjunlee/dev-relay --skill relay-merge
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/sungjunlee/dev-relay --skill relay-merge

简介

relay-merge 用于查找、检索和筛选相关信息。

  • 适合在 Codex、Claude、Cursor、Gemini CLI 中根据关键词或任务场景快速定位候选结果。
  • 通过 npx skills add 命令从指定 GitHub 仓库安装并使用。
  • 安装前需确认权限范围、维护状态,注意是否触发联网、命令执行或文件读写操作。
  • 适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

SKILL.md

Relay Merge

Explicitly merge a ready-to-merge PR and close the loop. Requires relay-review PR comment.

Process

0. Gate check — verify relay-review completed

${CLAUDE_SKILL_DIR}/scripts/gate-check.js $PR_NUM
  • Exit 0 (LGTM) → PR is ready to merge; proceed only if the user wants to land it now
  • Exit 1 (no comment) → STOP. Run relay-review first
  • Exit 1 (stale LGTM) → STOP. Run relay-review again for the latest commit
  • Exit 1 (CHANGES_REQUESTED) → STOP. Re-dispatch or fix the branch first
  • Exit 1 (ESCALATED) → STOP. Show unresolved issues to user

Intentional skip (hotfix, manual PR, trivial change):

${CLAUDE_SKILL_DIR}/scripts/gate-check.js $PR_NUM --skip "reason here"

This writes a <!-- relay-review-skip --> comment to the PR — maintaining audit trail even when review is bypassed. The skip reason is recorded on the PR for future reference. gate-check.js --skip does not invoke any executor or reviewer, so it does not consume manifest model_hints.

Do NOT merge without running gate-check. This is the audit trail that review actually happened (or was intentionally skipped with documented reason).

1. Merge + finalize cleanup

RUN_ID=<run-id-from-dispatch>
node ${CLAUDE_SKILL_DIR}/scripts/finalize-run.js --repo . --run-id "$RUN_ID" --merge-method squash --json

This script:

  • re-checks the latest PR audit trail and blocks merge if review.last_reviewed_sha is stale for the current HEAD
  • merges the PR and only advances the manifest after GitHub reports the PR as MERGED
  • best-effort deletes the remote branch after the merge is confirmed
  • marks the manifest merged
  • best-effort closes the linked issue
  • removes the retained worktree, deletes the local merged branch, and runs git worktree prune
  • records cleanup.status in the manifest

If the retained worktree is dirty, merge still succeeds but cleanup is recorded as failed and the manifest moves to next_action=manual_cleanup_required.

Emergency escape hatch:

node ${CLAUDE_SKILL_DIR}/scripts/finalize-run.js --repo . --run-id "$RUN_ID" --skip-review "hotfix" --json

finalize-run.js --skip-review bypasses reviewer invocation, so model_hints.review is a non-consumer on that path.

Operator-only force finalize for non-ready runs

node ${CLAUDE_SKILL_DIR}/scripts/finalize-run.js --repo . --run-id "$RUN_ID" \
  --force-finalize-nonready --reason "reviewer-swap exhausted, diff clean per manual inspection" --json

Use this only when an operator has independently checked that the PR is mergeable but the manifest cannot reach ready_to_merge. Typical cases: state stuck at escalated with a clean diff; reviewer-swap unavailable; manifest/PR state desync. This path is loud on purpose: it records a force_finalize event before merge and writes last_force into the manifest. --force-finalize-nonready requires --reason <non-empty-text>. --dry-run is observation-only on this path: it does not append force_finalize.

Do not use it for retry loops. Do not use it as a test shortcut. Do not use it to paper over a wrong manifest state that should be repaired instead.

Audit every use:

jq 'select(.event == "force_finalize")' ~/.relay/runs/<repo-slug>/<run-id>/events.jsonl

Bootstrap artifact reconciliation

When a run predates an artifact writer that the run itself introduced, use the structured reconciliation command instead of encoding that fact in a force-finalize reason:

node ${CLAUDE_SKILL_DIR}/scripts/relay-reconcile-artifact.js --repo . --run-id "$RUN_ID" \
  --artifact-path "~/.relay/runs/<repo-slug>/<run-id>/execution-evidence.json" \
  --writer-pr 267 --reason "run predates the artifact writer" --json

This stamps bootstrap_exempt in the manifest, emits force_finalize with bootstrap_exempt: true, and marks the run merged without invoking the PR merge path.

2. Sprint file update (if available)

If backlog/sprints/ has an active sprint file, update it. If no sprint file exists, skip this step.

Plan section — mark completed (was [~] during review):

- [x] #38 OAuth2 flow → PR #87 (merged)

Progress section — structured log entry with review round count:

- 2026-03-25 10:50: #38 dispatched → PR #87 → reviewed (LGTM, round 1) → merged

Running Context section — capture learnings for remaining tasks:

- OAuth2: PKCE flow using jose library. Tokens in httpOnly cookies.

3. Follow-up (if needed)

gh issue create --title "Follow-up: ..." --body "..."

Task file cleanup (move to backlog/completed/) happens at sprint end, not per-issue.

Sprint File State Transitions

[ ] #N Task name                          ← not started
[~] #N Task name → PR #M (reviewing)     ← dispatched, review in progress
[x] #N Task name → PR #M (merged)        ← completed

Sprint File Updates Summary

SectionWhat to updateWhen
Plan[~][x] with PR refEvery merge
ProgressStructured log with review roundsEvery merge
Running ContextLearnings that affect later tasksWhen something was discovered
Follow-up issuesNew GitHub issuesWhen review found out-of-scope work

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

37.11%
按下载量换算39

Claude

28.07%
按下载量换算30

Cursor

18.27%
按下载量换算19

Gemini CLI

10.19%
按下载量换算11

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills