Back to Details

workbench-waking-up

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Workbench Waking-Up

Workbench 唤醒流程

workbench-waking-up
is the context-restore protocol for the Multica Ultimate Workbench.
It pairs with closeout and synthesis work: close the loop with durable evidence, then wake the system by re-reading memory leads and live state before choosing what to do next.
Core line:
text
Restore context. Verify reality. Pick the next move.
workbench-waking-up
是Multica Ultimate Workbench的上下文恢复协议。
它与收尾总结工作相辅相成：先用可靠证据闭环，再通过重新读取记忆线索和实时状态唤醒系统，之后再选择下一步行动。
核心准则：
text
恢复上下文。验证实际状态。选择下一步行动。

When To Use

使用场景

Use this skill when the operator says or implies:
  • gm
    , 
    maxxedhere
    , 
    where are we
    , 
    what can move now
    ;
  • "最近干了什么", "我们能干什么", "有没有落下没续的";
  • "看看最近的记忆", "没聊的时间里更新了什么";
  • a direct-chat discovery should become visible to the wider Workbench;
  • a session handoff, cron run, Multica issue, repo branch, or runner state may be stale.
Default to the current Workbench repo and project when the target is obvious. Ask only when there are multiple plausible workspaces and the tool calls would be materially different.
当操作员说出或暗示以下内容时，使用本技能：
  • gm
    , 
    maxxedhere
    , 
    where are we
    , 
    what can move now
    ;
  • "最近干了什么", "我们能干什么", "有没有落下没续的";
  • "看看最近的记忆", "没聊的时间里更新了什么";
  • 私聊发现的信息需要同步至整个Workbench；
  • 会话交接、定时任务（cron）运行、Multica工单、仓库分支或运行器状态可能已过期。
当目标明确时，默认使用当前Workbench仓库和项目。只有当存在多个可行工作区且工具调用会产生实质性差异时，才需要询问确认。

Output Contract

输出规范

Return this shape unless the caller asks for a different format:
text
GM, operator.

TOPLINE
- state judgment:
- confidence: high | medium | low
- main quest:
- biggest drift:

WHAT CHANGED WHILE YOU WERE AWAY
WE'RE SO BACK
- confirmed landed / verified items
LET HIM COOK
- active runs, agents, cron jobs, or dogfoods
IT'S SO OVER
- blockers, failed jobs, or evidence gaps
DO NOT COOK
- parked, unapproved, dangerous, or explicitly deferred paths

REALITY CHECK
- repo/git:
- automation/cron:
- issues / PRs:
- knowledge/vault:
- runners/skills:

OPEN LOOPS
P0 — must resolve before new work
- ...
P1 — high leverage next
- ...
P2 — useful but not blocking
- ...
PARKED — intentionally not now
- ...

DROPPED THREADS / NOT YET RESUMED
- ...

ACTION MENU
1. ...
2. ...
3. ...

RECEIPTS
- commands, issue IDs, commit IDs, session recalls, and uncertainty labels
Use the GM labels lightly. They are shorthand for state, not a substitute for proof.
除非调用者要求其他格式，否则返回以下结构：
text
GM，操作员。

核心概览
- 状态判断：
- 置信度：高 | 中 | 低
- 核心任务：
- 最大偏差：

你离开期间的变化
已完成事项
- 已确认落地/已验证的条目
进行中事项
- 活跃的运行任务、Agent、定时任务或内部测试（dogfoods）
已终止事项
- 阻塞项、失败任务或证据缺口
暂停事项
- 已搁置、未批准、有风险或明确延期的路径

实际状态核查
- 仓库/git：
- 自动化/定时任务：
- 工单/PR：
- 知识库/ vault：
- 运行器/技能：

未闭环事项
P0 — 开展新工作前必须解决
- ...
P1 — 高优先级下一步
- ...
P2 — 有用但不阻塞
- ...
已搁置 — 暂不处理
- ...

未跟进的线索/尚未恢复的任务
- ...

操作菜单
1. ...
2. ...
3. ...

证据来源
- 命令、工单ID、提交ID、会话记录、不确定性标签
GM标签请谨慎使用，它们是状态的简写，不能替代实际证据。

Evidence Labels

证据标签

Every material claim should be traceable to one of these labels:
  • LIVE_VERIFIED
    — checked against current repo, issue, cron, CI, runner, or browser state.
  • CONFIRMED
    — backed by a durable source file, decision, issue comment, or committed artifact.
  • MEMORY_ONLY
    — recalled from session memory but not yet live-verified.
  • FLAGGED
    — evidence exists but is inconsistent, partial, or blocked.
  • STALE
    — prior claim is contradicted by current state or has aged out.
Memory is a lead, not truth. Say 
memory suggests
until live readback confirms it.
所有重要结论都必须可追溯至以下标签之一：
  • LIVE_VERIFIED
    — 已核对当前仓库、工单、定时任务、CI、运行器或浏览器状态。
  • CONFIRMED
    — 有可靠源文件、决策记录、工单评论或已提交工件作为支撑。
  • MEMORY_ONLY
    — 仅来自会话记忆，尚未经过实时验证。
  • FLAGGED
    — 存在证据但内容不一致、不完整或被阻塞。
  • STALE
    — 先前结论与当前状态矛盾或已过期。
记忆只是线索，而非事实。在实时反馈确认前，请注明“记忆显示”。

Wake Procedure

唤醒流程

Fast Wake

快速唤醒

Use for short orientation prompts.
  1. Recall recent sessions or the latest closeout summary.
  2. Check current repo root, branch, 
    git status --short --branch
    , and recent commits.
  3. Check active cron/jobs and obvious Multica issue state if available.
  4. Return the top drift and no more than three next actions.
适用于简短的定位需求。
  1. 调用近期会话或最新的收尾总结。
  2. 检查当前仓库根目录、分支、
    git status --short --branch
    以及近期提交记录。
  3. 若可用，检查活跃的定时任务/作业以及明显的Multica工单状态。
  4. 返回最主要的偏差信息，以及不超过三个下一步行动。

Standard Wake

标准唤醒

Default mode.
  1. Search memory for current project, recent open-loop terms, and known runner names.
  2. Verify repo state: branch, ahead/behind, dirty files, untracked evidence, recent commits, and remote freshness.
  3. Verify live Workbench surfaces: Multica issue status/runs/comments, cron jobs, autopilots, Research Vault or Sanity context if relevant.
  4. Verify runner health only when it changes the recommendation: Hermes, Codex, 
    ult-evo
    , Capy, MCP, browser/deploy, or remote cell.
  5. Collapse findings into P0/P1/P2/PARKED and a short action menu.
默认模式。
  1. 在记忆中搜索当前项目、近期未闭环事项的关键词以及已知运行器名称。
  2. 验证仓库状态：分支、领先/落后情况、未提交文件、未追踪证据、近期提交记录以及远程仓库新鲜度。
  3. 验证Workbench实时状态：Multica工单的状态/运行记录/评论、定时任务、自动导航（autopilots）、Research Vault或Sanity相关上下文。
  4. 仅当运行器健康状况会影响建议时，才验证其状态：Hermes、Codex、
    ult-evo
    、Capy、MCP、浏览器/部署或远程单元。
  5. 将结果整理为P0/P1/P2/已搁置事项，并生成简短操作菜单。

Deep Wake

深度唤醒

Use for a full operating review or multi-repo reset. Add targeted issue run messages, GitHub PR/CI/deploy readbacks, Capy/browser evidence, Research Vault queries, and sanitized screenshots or artifacts when needed.
适用于全面运行审查或多仓库重置。必要时添加针对性工单运行消息、GitHub PR/CI/部署反馈、Capy/浏览器证据、Research Vault查询结果，以及经过脱敏的截图或工件。

Drift-First Rule

偏差优先规则

Report drift before proposing new work. P0 drift includes:
  • branch ahead/behind or dirty baseline;
  • untracked artifacts that might be evidence or noise;
  • cron/autopilot fired but downstream delivery failed;
  • issue state disagrees with run state;
  • done
    , 
    merged
    , 
    deployed
    , or 
    pushed
    claim lacks live readback;
  • Research Vault, Sanity, or runner state contradicts the claimed source;
  • operator decision is still blocking a parent lane.
Do not bury drift under a motivational recap.
在提出新工作之前，先报告偏差信息。P0级偏差包括：
  • 分支领先/落后或基线未提交；
  • 未追踪的工件（可能是证据或无效信息）；
  • 定时任务/自动导航已触发，但下游交付失败；
  • 工单状态与运行状态不一致；
  • “完成”“合并”“部署”或“推送”的结论缺乏实时反馈验证；
  • Research Vault、Sanity或运行器状态与声称的来源矛盾；
  • 操作员的决策仍在阻塞主流程。
不要将偏差信息隐藏在激励性总结之下。

Session-To-Workbench Bridge

会话至Workbench的同步机制

When a direct chat produces a reusable behavior, packaging direction, or policy change, make it Workbench-visible instead of leaving it hidden in model memory. Pick the smallest durable surface:
Durable surfaceUse when
skills/workbench-*/SKILL.md
the behavior should be reusable by agents
AGENTS.md
every agent needs a routing or safety rule
SYNTHESIS.md
the operating model or strategy changed
DECISIONS.md
the rationale must survive disagreement
skills/README.md
the installable skill map changed
issue template or autopilot specfuture tasks need a reusable entry point
sanitized Multica issue/commentlive operators need immediate coordination
Do not store secrets, raw transcripts, private screenshots, request payloads, or live-only credentials. Store sanitized summaries and pointers only.
当私聊产生可复用行为、打包方向或政策变更时，需将其同步至Workbench可见位置，而非仅存于模型记忆中。选择最小的可靠存储位置：
可靠存储位置使用场景
skills/workbench-*/SKILL.md
该行为需被Agent复用
AGENTS.md
所有Agent都需要路由或安全规则
SYNTHESIS.md
运营模型或策略发生变更
DECISIONS.md
决策依据需保留以应对分歧
skills/README.md
可安装技能的映射关系发生变更
工单模板或自动导航规范未来任务需要可复用的入口
脱敏后的Multica工单/评论现场操作员需要即时协作
请勿存储机密信息、原始对话记录、私人截图、请求负载或仅适用于实时环境的凭证。仅存储脱敏总结和指向性信息。

Priority Rules

优先级规则

  • P0: sync drift, failed automation, blocked operator decision, empty or contradictory run output, missing proof for claimed completion.
  • P1: high-leverage review, dogfood gate, bounded issue update, or single synthesis step that unblocks a parent lane.
  • P2: useful cleanup, future packaging, optional product/design slices.
  • PARKED: explicitly deferred or unapproved routes. Do not revive them without new evidence and operator approval.
  • P0：同步偏差信息、失败的自动化任务、阻塞性的操作员决策、为空或矛盾的运行输出、声称已完成但缺少证据的事项。
  • P1：高价值审查、内部测试关卡、有边界的工单更新、或能解锁主流程的单步总结工作。
  • P2：有用的清理工作、未来打包任务、可选的产品/设计切片。
  • 已搁置：明确延期或未批准的路径。若无新证据和操作员批准，请勿重启。

Closeout Checklist

收尾检查清单

Before sending a wake report or bridge patch:
  • Memory leads checked or explicitly unavailable.
  • Live repo/issue/automation state checked when relevant.
  • Drift appears before new ideas.
  • Claims are evidence-labeled.
  • Action menu is capped at three to five moves.
  • Parked decisions remain parked.
  • Public docs contain no secrets, raw transcripts, live-only IDs, or private payloads.
发送唤醒报告或同步补丁前，请确认：
  • 已检查记忆线索，或明确表示无法获取记忆线索。
  • 相关的仓库/工单/自动化任务实时状态已检查。
  • 偏差信息在新想法之前展示。
  • 结论带有证据标签。
  • 操作菜单限制为3-5个选项。
  • 已搁置的决策仍保持搁置状态。
  • 公开文档中不包含机密信息、原始对话记录、仅适用于实时环境的ID或私人负载。