Adonis0123/adonis-skills / skills/agentic-review-handoff
agentic-review-handoff
VerifiedValidate pasted review findings before fixes; run same-session automatic Git review-fix-re-review with a headless Reviewer and no redundant confirmation; start fresh-eyes Git diff Review Intake; resume review-loop sessions or packets; get a DecisionConsult from another AI; or run first-principles/DDD/high-cohesion review. Requires Git and review/consult intent; ordinary implementation or commit-only requests do not start a review loop.
evals/evals.json
View on GitHub{
"skill_name": "agentic-review-handoff",
"description": "Auto/classic/direct Grok routing, no-confirm completion, truthful Intake/Handoff origin, protocol lifecycle, and trigger regressions. Historical trigger measurements are not evidence for v3.8.0.",
"trigger_evals": [
{
"id": "pos-auto-loop-zh",
"query": "我在 adonis-skills 这个 git 仓库刚改完 review-loop,想在当前会话里直接开 auto loop:用 codex 当 headless reviewer,自己修到 PASS 或预算耗尽,中间别问我。帮我跑 review→fix→re-review 闭环。",
"should_trigger": true
},
{
"id": "pos-dual-ai-session",
"query": "same session dual AI review closed loop please — visible me as fixer, headless grok as reviewer on the uncommitted diff under ~/coding/my-app, zero mid-loop human interrupts until terminal report.",
"should_trigger": true
},
{
"id": "pos-review-fix-rereview",
"query": "能不能自动 review-fix-re-review 三轮?仓库是 ~/work/breeze,reviewer 用 claude,先审当前 working tree,BLOCKED 了我改完再 continue,别让我两边窗口来回点。",
"should_trigger": true
},
{
"id": "pos-consult",
"query": "决策咨询:认证该用 session cookie 还是 JWT?把题框定好后用 review-loop consult 问一下 peer=codex 的立场,advisory 就行,别静默拍板。repo 在 ~/coding/auth-svc。",
"should_trigger": true
},
{
"id": "pos-sessions-resume",
"query": "上次 auto loop 用的 headless codex reviewer 会话在 Desktop 列表里找不到了。给我 codex/grok/claude 恢复对话的命令,或者直接 list review-loop sessions。仓库根是 /Users/you/coding/adonis-skills。",
"should_trigger": true
},
{
"id": "pos-second-eyes-packet",
"query": "帮我当 second pair of eyes 审一下 feat/cart-patch 上 staged 的 apps/api/cart-handler.ts,按 agentic packet 写 findings,有阻塞就整理 fix handoff 给后续修复。",
"should_trigger": true
},
{
"id": "pos-continue-packet",
"query": ".review-handoff/active 里最新 packet 是 BLOCKED 且我已写好 Fix Completion,请 run --continue 进入 re-review。repo: ~/coding/payment-svc 分支 feat/validator。",
"should_trigger": true
},
{
"id": "pos-pwc-fix-it",
"query": "上轮 re-review 是 PASS_WITH_CONCERNS,lifecycle 还在 awaiting_user_decision。我说修一下那两个命名 concern,按 packet 协议续到 round 2。",
"should_trigger": true
},
{
"id": "pos-feedback-validate",
"query": "团队贴了 review 反馈说 auth.ts 空 token 会 500,请当缺陷报告验证后再整理 validated findings / fix handoff,不要把反馈当 ground truth。仓库在 monorepo 根。",
"should_trigger": true
},
{
"id": "pos-first-principles-deep",
"query": "对 packages/billing 这批改动做 first-principles + DDD 边界审查,跨模块耦合风险也看一下,产出带 Verdict 的 review packet。",
"should_trigger": true
},
{
"id": "pos-architecture-hardening-nested-review",
"query": "architecture-hardening-loop 已完成扫描和证据分诊,现在按它的硬依赖显式调用 agentic-review-handoff,用 Grok 对已修复的冻结范围跑 review-fix-re-review;不要重跑架构 scanner。",
"should_trigger": true
},
{
"id": "neg-ordinary-impl",
"query": "帮我在 apps/web 加一个 React 登录表单组件,用 Zod 校验 email/password,写好后 commit 就行,不需要 review。",
"should_trigger": false
},
{
"id": "neg-generic-staged-review",
"query": "简单看看 git status 里 staged 的 diff 有没有明显 typo,口头说两句就行,不用写 packet 也不用开 loop。",
"should_trigger": false
},
{
"id": "neg-non-git-folder",
"query": "我在 ~/Downloads/notes-scratch(不是 git 仓库)里有几个 markdown 草稿,帮我润色标题措辞。",
"should_trigger": false
},
{
"id": "neg-named-other-skill",
"query": "用 weekly-report skill 生成我这周的工作周报,仓库列表是 adonis-skills 和 breeze,不要走 review handoff。",
"should_trigger": false
},
{
"id": "neg-codex-plugin-once",
"query": "在 Claude Code 里直接 /codex:review 一下当前 uncommitted 就够了,不要 agentic-review-handoff,也不要 auto loop。",
"should_trigger": false
},
{
"id": "neg-copy-edit-comments",
"query": "帮我把 PR 评论里这几句英文 review comment 润色得更礼貌,纯文案修改,不涉及代码修复闭环。",
"should_trigger": false
},
{
"id": "neg-commit-only",
"query": "改完了,按 conventional commits 帮我 commit 当前 staged 变更,message 用 feat(auth): ...",
"should_trigger": false
},
{
"id": "neg-unit-test-impl",
"query": "给 cart-handler.ts 的 PATCH 路由补三个 vitest 单测(含 quantity=0),实现完跑测试绿即可。",
"should_trigger": false
},
{
"id": "neg-named-grok-builtin-review",
"query": "用 Grok Build 内置的 /review 命令看一下当前 uncommitted,不要 agentic-review-handoff 也不要写 .review-handoff packet。",
"should_trigger": false
},
{
"id": "neg-prompt-only-composer",
"query": "帮我写一个 prompt,让另一个 AI review 当前 staged diff;只生成可复制提示词,不要现在执行 review、不要开 packet loop。",
"should_trigger": false
},
{
"id": "neg-architecture-hardening-parent",
"query": "把 packages/billing 交给 architecture-hardening-loop:先独立扫描架构问题,证据分诊后再做最小修复和测试,Grok review 每一批,然后同范围复扫直到没有可修项。这是 parent scoped harden,不要开一次性 review-fix-re-review packet。",
"should_trigger": false
}
],
"evals": [
{
"id": 1,
"name": "scenario-A-auto-loop-start",
"prompt": "在 git 仓库 ${MOCK_REPO}(分支 feat/cart-patch)对 staged 的 apps/api/cart-handler.ts 发起 auto loop:review-loop run --reviewer=codex,中间零人工,直到终态。diff 摘要:PATCH /api/cart 缺 quantity=0 边界。",
"expected_output": "无可验证 implementer context,调用/说明 review-loop run --intake(非 open/bind),packet 首个 H1 是 # Review Intake;钉死 base;冻结 evidence;headless reviewer;BLOCKED 时有 Fix Handoff 或 blocked 状态机出口。",
"files": []
},
{
"id": 2,
"name": "scenario-B-continue-after-fix",
"prompt": "${MOCK_REPO}/.review-handoff/active/ 最新 packet last_anchor=fix_completion,上轮 BLOCKED。请 run --continue 做 re-review。",
"expected_output": "不创建新 packet;run --continue;re-review schema(Prior reassessment + New Findings 全列表 + Regression Surface + Verdict);更新 frontmatter/round。",
"files": []
},
{
"id": 3,
"name": "scenario-C-consult",
"prompt": "在 ${MOCK_REPO} 上 明确运行 review-loop consult --peer=grok:要不要把 cart 校验抽到 shared package?advisory,不要静默采纳。",
"expected_output": "review-loop consult;记录到 .review-handoff/runtime/consults/;明确 advisory 边界;不加载 auto-loop-contract 或 classic packet references。",
"files": []
},
{
"id": 4,
"name": "scenario-D-sessions-resume",
"prompt": "给我 codex 恢复上次 headless reviewer 会话的命令。仓库 ${MOCK_REPO}。",
"expected_output": "review-loop sessions(或等价 list);输出 copy-ready resume 命令;不调用已删除 dual-window;不加载任何 reference。",
"files": []
},
{
"id": 5,
"name": "scenario-E-classic-reviewer-initiated",
"prompt": "我刚改了 apps/api/cart-handler.ts staged 但未 commit。帮 review 并整理 fix handoff。仓库根 ${MOCK_REPO},分支 feat/cart-patch。缺 quantity=0 边界。",
"expected_output": "packet 含 # Review Intake → # Review Findings →(若 BLOCKED)# Fix Handoff;不用 dual-window 命令。",
"files": []
},
{
"id": 6,
"name": "scenario-F-pwc-fix-it",
"prompt": "packet 末段 # Re-review Verdict=PASS_WITH_CONCERNS,lifecycle=awaiting_user_decision。用户说修一下。${MOCK_REPO}。",
"expected_output": "续同一 packet 进入 fix / round+1;不强制 archive;不新建 packet。",
"files": []
},
{
"id": 7,
"name": "scenario-G-first-pass-pass-archive",
"prompt": "只 rename const DATE_FMT_ISO → DATE_FORMAT_ISO_8601,请 review。${MOCK_REPO} 分支 feat/rename-const。",
"expected_output": "这是无可验证 implementer context 的 fresh-eyes review,走 classic intake;packet 从 # Review Intake 开始。PASS/NO_FINDINGS 时无 Fix Handoff 并归档。",
"files": []
},
{
"id": 8,
"name": "scenario-H-subdir-cwd",
"prompt": "cwd 在 ${MOCK_REPO}/apps/web/src/components/auth/,对 staged RegisterForm 密码校验改动做 review。",
"expected_output": "无 implementer context 时走 classic intake;git rev-parse --show-toplevel → packet 只写在 repo root .review-handoff/,不在 apps/web 下。",
"files": []
},
{
"id": 9,
"name": "scenario-I-no-dual-window",
"prompt": "用户误说 open/bind dual-window 审查。${MOCK_REPO}。",
"expected_output": "拒绝/迁移提示:T8 已删 dual-window,改用 review-loop run / fix-completion / consult。",
"files": []
},
{
"id": 10,
"name": "scenario-J-budget-terminal-report",
"prompt": "auto loop 终态 budget_exhausted,请给用户终态汇报。${MOCK_REPO}。",
"expected_output": "status=budget_exhausted、双方立场/未决项、推荐 --rounds +N;不调用 board/summary。",
"files": []
},
{
"id": 11,
"name": "scenario-K-healthy-silent-reviewer",
"prompt": "Grok reviewer 已经 5 分钟没有 stdout,但 review-loop 的 progress 仍显示进程 active,deadline 还没到。现在是不是应该 kill 后重试?${MOCK_REPO}。",
"expected_output": "继续等待同一 Reviewer;stdout 静默不是失败,不手工 kill、不重试、不启动第二个 Reviewer。只在 STOP、进程退出或 adapter deadline 到达后按 DELIVERY_UNKNOWN 处理。",
"files": []
},
{
"id": 12,
"name": "scenario-L-same-session-implementer-auto",
"prompt": "你刚在当前会话实现并验证了 apps/api/cart-handler.ts。现在继续做 review-fix-re-review 到 PASS,中途不用问我。仓库 ${MOCK_REPO}。",
"expected_output": "当前会话有可验证 implementer context 且用户明确要求闭环,走默认 review-loop run auto;允许创建 implementer Review Handoff,不添加 --intake。默认 completion=pass。Reviewer 未指定时启动前用提问工具问一次,推荐与当前宿主不同的产品,之后不再问;PASS_WITH_CONCERNS 自动修复续审,不询问用户。",
"files": []
},
{
"id": 13,
"name": "scenario-M-source-prompt-v2-digest-mismatch",
"prompt": "反馈回显了一个 format_version 2 Review-Prompt-ID,但 prompt 的 --verify-prompt 现在返回 scope_digest_mismatch;HEAD 和 status 路径仍相同。请开始 review。${MOCK_REPO}。",
"expected_output": "把 source prompt 判为 stale,停止按旧 prompt 开审并要求重新生成/明确新范围;不因 HEAD/status 路径相同继续,不 backfill digest,也不把 prompt 摘要当当前证据。",
"files": []
},
{
"id": 14,
"name": "scenario-N-default-pwc-needs-no-confirmation",
"prompt": "显式调用 agentic-review-handoff 跑当前 diff,没有指定 reviewer。首轮返回 PASS_WITH_CONCERNS,两个 concern 都在冻结范围内且可逆,预算还有两轮。不要真的运行 CLI,只说明默认状态机。${MOCK_REPO}。",
"expected_output": "启动前问一次 Reviewer(推荐非宿主产品;无提问工具时直接选非宿主并在报告里写明),completion=pass。返回 concerns_require_fix/blocked 和 Fix Handoff,Fixer 修复、验证、写 Fix Completion 后 run --continue;不进入 awaiting_user_decision,不问用户是否继续,也不自动接受 concern。",
"files": []
},
{
"id": 15,
"name": "proven-local-startup-fault",
"prompt": "用户已经授权发送 Grok 复审。CLI 已退出;本地 launcher 代码和 stderr 证明只读沙箱初始化因一条目标不存在的 socket 符号链接失败,尚未调用模型;可在现有权限内临时移开后恢复。结果标为 DELIVERY_UNKNOWN。继续完成复审。 ${MOCK_REPO}。",
"expected_output": "不把复审扩展为主机维修;不移动或删除 socket,不启动 Docker。保留账号与范围,报告所需复审被启动错误阻塞。仅纠正已授权的调用参数错误;不关闭用户明确配置的沙箱。",
"files": []
},
{
"id": 16,
"name": "ambiguous-delivery-with-socket-error",
"prompt": "Grok 运行五分钟后连接中断,同时 stderr 提到 Docker socket,无法确认模型是否已收到请求。现在继续处理。 ${MOCK_REPO}。",
"expected_output": "先核实既有会话状态,不因 socket 关键词假定未发送;不盲目重试,不禁用沙箱,不改 live socket。无法查明时报告真实阻塞。",
"files": []
},
{
"id": 17,
"name": "recovery-path-replaced",
"prompt": "已确认启动前失败并完成一次临时移开失效链接的修复,复审完成;恢复时发现原路径被其他进程创建了有效 socket。完成收尾。 ${MOCK_REPO}。",
"expected_output": "不覆盖或删除有效 socket;保留备份,报告恢复冲突,区分复审结果和未完成收尾,不宣称已完整恢复。",
"files": []
},
{
"id": 18,
"name": "repeat-startup-fault",
"prompt": "已确认未发送,最小修复后重试一次仍在同一启动步骤失败。用户希望自动闭环。 ${MOCK_REPO}。",
"expected_output": "停止重试,保留诊断,报告阻塞;不无限重试、不重置预算、不转去修 Docker 或账号配置。",
"files": []
},
{
"id": 19,
"name": "direct-numbered-grok-review",
"prompt": "用 grok003 帮我看看 src/cart.ts 的未提交修改有没有问题,给我结论即可。仓库 ${MOCK_REPO}。",
"expected_output": "发现既有 grok003 入口,在其定义 shell 中传递 prompt-file;仅使用读取搜索工具,不强制 OS sandbox。不建立 packet/ledger,不修源码,不静默切账号。",
"files": []
},
{
"id": 20,
"name": "explicit-grok-auto-loop",
"prompt": "用 Grok 进行自动 review-fix-re-review,最多三轮;当前会话刚实现了修改。仓库 ${MOCK_REPO}。",
"expected_output": "保留 review-loop run --reviewer=grok --rounds 3,默认 Handoff,不误走单次直调。",
"files": []
},
{
"id": 21,
"name": "optional-grok-startup-failure",
"prompt": "问 Grok 这个方案的利弊。调用返回 could not apply read-only sandbox profile: /var/run/docker.sock endpoint is a symlink。主任务还有独立的本地测试可做。",
"expected_output": "普通咨询直调 Grok。失败后不修 Docker、不移动 socket、不改账号、不假报结论;记录 Grok 意见不可用,继续独立测试。",
"files": []
},
{
"id": 22,
"name": "reviewer-unspecified-asks-once",
"prompt": "我在 Codex 里刚改完 apps/api/cart-handler.ts,帮我跑 review-fix-re-review 到 PASS。仓库 ${MOCK_REPO}。没说用哪个 AI 审。",
"expected_output": "启动 run 前用提问工具问一次 Reviewer,推荐项不是 Codex(推荐 Grok 或 Claude);不静默默认 Codex,不在轮次中再问;拿到答案后传 --reviewer。",
"files": []
},
{
"id": 23,
"name": "named-account-reviewer",
"prompt": "用 grok002 这个账号当 reviewer 跑自动 review-fix-re-review,当前会话刚实现了改动。仓库 ${MOCK_REPO}。",
"expected_output": "不再询问 Reviewer;调用 review-loop run --reviewer=grok --reviewer-command=grok002,不复制凭据、不轮换账号;--continue 继承同一 launcher,sessions 输出 grok002 -r <id>。",
"files": []
},
{
"id": 24,
"name": "budget-exhausted-unattended-defers",
"prompt": "现在没人在电脑前,会话里不能提问。auto loop 第 3 轮终态 budget_exhausted,unresolved 还有 2 条。任务里还有一段不依赖这次改动的文档要更新。${MOCK_REPO}。",
"expected_output": "先判断无人值守(装了 sideby 就看 sideby unattended status --json 的 on,否则以宿主说明为准)。不提问、不自行加 --rounds、不再修剩余发现;记一条待用户决定(2 条发现、所在分支、packet;有 sideby 用 sideby unattended defer,没有就只写进终态报告);不提交、不合并带未决发现的代码;继续做不依赖它的文档更新;终态报告列出这一项。",
"files": []
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |